Update test_qlib_from_source.yml

Update test_qlib_from_pip.yml
2026-07-21 19:27:36 +08:00 · 2022-06-29 17:07:05 +08:00 · 2022-06-29 17:02:57 +08:00 · 2022-06-29 17:01:51 +08:00 · 2022-06-29 17:00:35 +08:00 · 2022-06-29 13:38:36 +08:00
51 changed files with 2052 additions and 621 deletions
--- a/.github/workflows/test_macos.yml
+++ b/.github/workflows/test_macos.yml
@@ -1,94 +0,0 @@
-# There are some issues (in the downloading data phase) on MacOS when running with other tests. So we split it into an individual config.
-name: Test MacOS
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-
-jobs:
-  build:
-
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [macos-11, macos-latest]
-        # not supporting 3.6 due to annotations is not supported https://stackoverflow.com/a/52890129
-        python-version: [3.7, 3.8]
-
-    steps:
-    - uses: actions/checkout@v2
-
-    - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v2
-      with:
-        python-version: ${{ matrix.python-version }}
-
-    - name: Lint with Black
-      run: |
-        cd ..
-        python -m pip install pip --upgrade
-        python -m pip install wheel --upgrade
-        python -m pip install black
-        python -m black qlib -l 120 --check --diff
-    # Test Qlib installed with pip
-
-    - name: Check Qlib with flake8
-      run: |
-        pip install --upgrade pip
-        pip install flake8
-        flake8 --ignore=E501,F541,E266,E402,W503,E731,E203 --per-file-ignores="__init__.py:F401,F403" qlib
-
-    - name: Install Qlib with pip
-      run: |
-          python -m pip install numpy==1.19.5
-          python -m pip install pyqlib --ignore-installed ruamel.yaml numpy
-    - name: Make html with sphnix
-      run: |
-        pip install -U sphinx
-        pip install sphinx_rtd_theme readthedocs_sphinx_ext
-        pip install --exists-action=w --no-cache-dir -r docs/requirements.txt 
-        cd docs 
-        sphinx-build -b html . build
-        cd ..
-    - name: Install Lightgbm for MacOS
-      run: |
-        /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Microsoft/qlib/main/.github/brew_install.sh)"
-        HOMEBREW_NO_AUTO_UPDATE=1 brew install lightgbm
-        # FIX MacOS error: Segmentation fault
-        # reference: https://github.com/microsoft/LightGBM/issues/4229
-        wget https://raw.githubusercontent.com/Homebrew/homebrew-core/fb8323f2b170bd4ae97e1bac9bf3e2983af3fdb0/Formula/libomp.rb
-        brew unlink libomp
-        brew install libomp.rb
-    - name: Test data downloads
-      run: |
-        python scripts/get_data.py qlib_data --name qlib_data_simple --target_dir ~/.qlib/qlib_data/cn_data_simple --interval 1d --region cn
-        python -c "import os; userpath=os.path.expanduser('~'); os.rename(userpath + '/.qlib/qlib_data/cn_data_simple', userpath + '/.qlib/qlib_data/cn_data')"
-        azcopy copy https://qlibpublic.blob.core.windows.net/data /tmp/qlibpublic --recursive
-        mv /tmp/qlibpublic/data tests/.data
-    - name: Test workflow by config (install from pip)
-      run: |
-        python qlib/workflow/cli.py examples/benchmarks/LightGBM/workflow_config_lightgbm_Alpha158.yaml
-        python -m pip uninstall -y pyqlib
-    # Test Qlib installed from source
-    - name: Install Qlib from source
-      run: |
-        python -m pip install --upgrade cython
-        python -m pip install numpy jupyter jupyter_contrib_nbextensions
-        python -m pip install -U scipy scikit-learn # installing without this line will cause errors on GitHub Actions, while instsalling locally won't
-        python -m pip install gym tianshou torch
-        pip install -e .
-    - name: Install test dependencies
-      run: |
-        python -m pip install --upgrade pip
-        python -m pip install -U pyopenssl idna
-        python -m pip install black pytest
-    - name: Unit tests with Pytest
-      run: |
-        pip install -r scripts/data_collector/pit/requirements.txt
-        cd tests
-        python -m pytest . --durations=0
-    - name: Test workflow by config (install from source)
-      run: |
-          python qlib/workflow/cli.py examples/benchmarks/LightGBM/workflow_config_lightgbm_Alpha158.yaml
--- a/.github/workflows/test_qlib_from_pip.yml
+++ b/.github/workflows/test_qlib_from_pip.yml
@@ -0,0 +1,57 @@
+name: Test qlib from pip
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  build:
+    timeout-minutes: 120
+
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [windows-latest, ubuntu-18.04, ubuntu-20.04, macos-11, macos-latest]
+        # not supporting 3.6 due to annotations is not supported https://stackoverflow.com/a/52890129
+        python-version: [3.7, 3.8]
+
+    steps:
+    - name: Test qlib from pip
+      uses: actions/checkout@v2
+
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+
+    - name: Update pip to the latest version
+      run: |
+        python -m pip install --upgrade pip
+
+    - name: Qlib installation test
+      run: |
+        python -m pip install pyqlib
+        # Specify the numpy version because the numpy upgrade caused the CI test to fail, 
+        # and this line of code will be removed when the next version of qlib is released.
+        python -m pip install "numpy<1.23"
+
+    - name: Install Lightgbm for MacOS
+      if: ${{ matrix.os == 'macos-11' || matrix.os == 'macos-latest' }}
+      run: |
+        /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Microsoft/qlib/main/.github/brew_install.sh)"
+        HOMEBREW_NO_AUTO_UPDATE=1 brew install lightgbm
+        # FIX MacOS error: Segmentation fault
+        # reference: https://github.com/microsoft/LightGBM/issues/4229
+        wget https://raw.githubusercontent.com/Homebrew/homebrew-core/fb8323f2b170bd4ae97e1bac9bf3e2983af3fdb0/Formula/libomp.rb
+        brew unlink libomp
+        brew install libomp.rb
+
+    - name: Downloads dependencies data
+      run: |
+        python scripts/get_data.py qlib_data --name qlib_data_simple --target_dir ~/.qlib/qlib_data/cn_data --interval 1d --region cn
+
+    - name: Test workflow by config
+      run: |
+        qrun examples/benchmarks/LightGBM/workflow_config_lightgbm_Alpha158.yaml
--- a/.github/workflows/test_qlib_from_source.yml
+++ b/.github/workflows/test_qlib_from_source.yml
@@ -1,4 +1,4 @@
-name: Test
+name: Test qlib from source

 on:
  push:
@@ -8,42 +8,60 @@ on:

 jobs:
  build:
+    timeout-minutes: 120

    runs-on: ${{ matrix.os }}
    strategy:
      matrix:
-        os: [windows-latest, ubuntu-18.04, ubuntu-20.04]
+        os: [windows-latest, ubuntu-18.04, ubuntu-20.04, macos-11, macos-latest]
        # not supporting 3.6 due to annotations is not supported https://stackoverflow.com/a/52890129
        python-version: [3.7, 3.8]

    steps:
-    - uses: actions/checkout@v2
+    - name: Test qlib from source
+      uses: actions/checkout@v2

    - name: Set up Python ${{ matrix.python-version }}
      uses: actions/setup-python@v2
      with:
        python-version: ${{ matrix.python-version }}

+    - name: Update pip to the latest version
+      run: |
+        python -m pip install --upgrade pip
+
+    - name: Installing pytorch for macos
+      if: ${{ matrix.os == 'macos-11' || matrix.os == 'macos-latest' }}
+      run: |
+        python -m pip install torch torchvision torchaudio
+
+    - name: Installing pytorch for ubuntu
+      if: ${{ matrix.os == 'ubuntu-18.04' || matrix.os == 'ubuntu-20.04' }}
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu
+
+    - name: Installing pytorch for windows
+      if: ${{ matrix.os == 'windows-latest' }}
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install torch torchvision torchaudio
+
+    - name: Set up Python tools
+      run: |
+        python -m pip install --upgrade cython
+        python -m pip install -e .[dev]
+
    - name: Lint with Black
      run: |
-        pip install --upgrade pip
-        pip install black wheel
-        black qlib -l 120 --check --diff
-
-    - name: Install Qlib with pip
-      run: |
-        pip install numpy==1.19.5 ruamel.yaml
-        pip install pyqlib --ignore-installed
+        black . -l 120 --check --diff

    - name: Make html with sphinx
      run: |
-        pip install -U sphinx
-        pip install sphinx_rtd_theme readthedocs_sphinx_ext
-        pip install --exists-action=w --no-cache-dir -r docs/requirements.txt 
        cd docs 
        sphinx-build -b html . build
        cd ..
-        
+
    # Check Qlib with pylint
    # TODO: These problems we will solve in the future. Important among them are: W0221, W0223, W0237, E1102
      # C0103: invalid-name
@@ -67,11 +85,9 @@ jobs:
      # W1309: f-string-without-interpolation
      # E1102: not-callable
      # E1136: unsubscriptable-object
-    # References for parameters: https://github.com/PyCQA/pylint/issues/4577#issuecomment-1000245962 
+    # References for parameters: https://github.com/PyCQA/pylint/issues/4577#issuecomment-1000245962
    - name: Check Qlib with pylint
      run: |
-        pip install --upgrade pip
-        pip install pylint
        pylint --disable=C0104,C0114,C0115,C0116,C0301,C0302,C0411,C0413,C1802,R0401,R0801,R0902,R0903,R0911,R0912,R0913,R0914,R0915,R1720,W0105,W0123,W0201,W0511,W0613,W1113,W1514,E0401,E1121,C0103,C0209,R0402,R1705,R1710,R1725,R1735,W0102,W0212,W0221,W0223,W0231,W0237,W0612,W0621,W0622,W0703,W1309,E1102,E1136 --const-rgx='[a-z_][a-z0-9_]{2,30}$' qlib --init-hook "import astroid; astroid.context.InferenceContext.max_inferred = 500"

    # The following flake8 error codes were ignored:
@@ -95,47 +111,40 @@ jobs:
        # Description: If there is whitespace before ":", it cannot pass the black check.
    - name: Check Qlib with flake8
      run: |
-        pip install --upgrade pip
-        pip install flake8
        flake8 --ignore=E501,F541,E266,E402,W503,E731,E203 --per-file-ignores="__init__.py:F401,F403" qlib

    # https://github.com/python/mypy/issues/10600
    - name: Check Qlib with mypy
      run: |
-        pip install mypy
        mypy qlib --install-types --non-interactive || true
-        mypy qlib
+        mypy qlib --verbose

    - name: Test data downloads
      run: |
-        python scripts/get_data.py qlib_data --name qlib_data_simple --target_dir ~/.qlib/qlib_data/cn_data_simple --interval 1d --region cn
-        python -c "import os; userpath=os.path.expanduser('~'); os.rename(userpath + '/.qlib/qlib_data/cn_data_simple', userpath + '/.qlib/qlib_data/cn_data')"
+        python scripts/get_data.py qlib_data --name qlib_data_simple --target_dir ~/.qlib/qlib_data/cn_data --interval 1d --region cn
        azcopy copy https://qlibpublic.blob.core.windows.net/data/rl /tmp/qlibpublic/data --recursive
        mv /tmp/qlibpublic/data tests/.data

-    - name: Test workflow by config (install from pip)
+    - name: Install Lightgbm for MacOS
+      if: ${{ matrix.os == 'macos-11' || matrix.os == 'macos-latest' }}
      run: |
-        python qlib/workflow/cli.py examples/benchmarks/LightGBM/workflow_config_lightgbm_Alpha158.yaml
-        python -m pip uninstall -y pyqlib
-
-     # Test Qlib installed from source
-    - name: Install Qlib from source
-      run: |
-        pip install --upgrade cython jupyter jupyter_contrib_nbextensions numpy scipy scikit-learn # installing without this line will cause errors on GitHub Actions, while instsalling locally won't
-        pip install gym tianshou torch
-        pip install -e .
-
-    - name: Install test dependencies
-      run: |
-        pip install --upgrade pip
-        pip install black pytest
-
-    - name: Unit tests with Pytest
-      run: |
-        pip install -r scripts/data_collector/pit/requirements.txt
-        cd tests
-        python -m pytest . --durations=10
+        /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Microsoft/qlib/main/.github/brew_install.sh)"
+        HOMEBREW_NO_AUTO_UPDATE=1 brew install lightgbm
+        # FIX MacOS error: Segmentation fault
+        # reference: https://github.com/microsoft/LightGBM/issues/4229
+        wget https://raw.githubusercontent.com/Homebrew/homebrew-core/fb8323f2b170bd4ae97e1bac9bf3e2983af3fdb0/Formula/libomp.rb
+        brew unlink libomp
+        brew install libomp.rb

    - name: Test workflow by config (install from source)
      run: |
+        # Version 0.52.0 of numba must be installed manually in CI, otherwise it will cause incompatibility with the latest version of numpy.
+        python -m pip install numba==0.52.0
+        # You must update numpy manually, because when installing python tools, it will try to uninstall numpy and cause CI to fail.
+        python -m pip install --upgrade numpy
        python qlib/workflow/cli.py examples/benchmarks/LightGBM/workflow_config_lightgbm_Alpha158.yaml
+
+    - name: Unit tests with Pytest
+      run: |
+        cd tests
+        python -m pytest . -m "not slow" --durations=0
--- a/.github/workflows/test_qlib_from_source_slow.yml
+++ b/.github/workflows/test_qlib_from_source_slow.yml
@@ -0,0 +1,56 @@
+name: Test qlib from source slow
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  build:
+    timeout-minutes: 120
+
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [windows-latest, ubuntu-18.04, ubuntu-20.04, macos-11, macos-latest]
+        # not supporting 3.6 due to annotations is not supported https://stackoverflow.com/a/52890129
+        python-version: [3.7, 3.8]
+
+    steps:
+    - name: Test qlib from source slow
+      uses: actions/checkout@v2
+
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+
+    - name: Set up Python tools
+      run: |
+        pip install --upgrade cython numpy pip
+        pip install -e .[dev]
+
+    - name: Downloads dependencies data
+      run: |
+        python scripts/get_data.py qlib_data --name qlib_data_simple --target_dir ~/.qlib/qlib_data/cn_data --interval 1d --region cn
+
+    - name: Install Lightgbm for MacOS
+      if: ${{ matrix.os == 'macos-11' || matrix.os == 'macos-latest' }}
+      run: |
+        /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Microsoft/qlib/main/.github/brew_install.sh)"
+        HOMEBREW_NO_AUTO_UPDATE=1 brew install lightgbm
+        # FIX MacOS error: Segmentation fault
+        # reference: https://github.com/microsoft/LightGBM/issues/4229
+        wget https://raw.githubusercontent.com/Homebrew/homebrew-core/fb8323f2b170bd4ae97e1bac9bf3e2983af3fdb0/Formula/libomp.rb
+        brew unlink libomp
+        brew install libomp.rb
+
+    - name: Unit tests with Pytest
+      uses: nick-fields/retry@v2
+      with:
+        timeout_minutes: 120
+        max_attempts: 3
+        command: |
+          cd tests
+          python -m pytest . -m "slow" --durations=0
--- a/.mypy.ini
+++ b/.mypy.ini
@@ -1,6 +1,6 @@
 [mypy]
 exclude = (?x)(
-    ^qlib/backtest
+    ^qlib/backtest/high_performance_ds\.py$
    | ^qlib/contrib
    | ^qlib/data
    | ^qlib/model
--- a/README.md
+++ b/README.md
@@ -458,7 +458,7 @@ Before we released Qlib as an open-source project on Github in Sep 2020, Qlib is

 This project welcomes contributions and suggestions.  
 **Here are some 
-[code standards](docs/developer/code_standard.rst) for submiting a pull request.**
+[code standards and development guidance](docs/developer/code_standard_and_dev_guide.rst) for submiting a pull request.**

 Making contributions is not a hard thing. Solving an issue(maybe just answering a question raised in [issues list](https://github.com/microsoft/qlib/issues) or [gitter](https://gitter.im/Microsoft/qlib)), fixing/issuing a bug, improving the documents and even fixing a typo are important contributions to Qlib.

--- a/docs/developer/code_standard_and_dev_guide.rst
+++ b/docs/developer/code_standard_and_dev_guide.rst
@@ -45,4 +45,16 @@ When you submit a PR request, you can check whether your code passes the CI test
 .. code-block:: bash

    pip install -e .[dev]
-    pre-commit install
+    pre-commit install
+
+
+=================================
+Development Guidance
+=================================
+
+As a developer, you often want make changes to `Qlib` and hope it would reflect directly in your environment without reinstalling it. You can install `Qlib` in editable mode with following command.
+The `[dev]` option will help you to install some related packages when developing `Qlib` (e.g. pytest, sphinx)
+
+.. code-block:: bash
+
+    pip install -e .[dev]
--- a/qlib/init.py
+++ b/qlib/init.py
@@ -2,7 +2,7 @@
 # Licensed under the MIT License.
 from pathlib import Path

-__version__ = "0.8.6"
+__version__ = "0.8.6.99"
 __version__bak = __version__  # This version is backup for QlibConfig.reset_qlib_version
 import os
 from typing import Union
--- a/qlib/backtest/init.py
+++ b/qlib/backtest/init.py
@@ -5,7 +5,7 @@ from __future__ import annotations

 import copy
 from pathlib import Path
-from typing import TYPE_CHECKING, Generator, List, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Any, Generator, List, Optional, Tuple, Union

 import pandas as pd

@@ -23,7 +23,6 @@ from ..utils import init_instance_by_config
 from .backtest import backtest_loop, collect_data_loop
 from .decision import Order
 from .exchange import Exchange
-from .position import Position
 from .utils import CommonInfrastructure

 # make import more user-friendly by adding `from qlib.backtest import STH`
@@ -44,7 +43,7 @@ def get_exchange(
    min_cost: float = 5.0,
    limit_threshold: Union[Tuple[str, str], float, None] = None,
    deal_price: Union[str, Tuple[str], List[str]] = None,
-    **kwargs,
+    **kwargs: Any,
 ) -> Exchange:
    """get_exchange

@@ -52,14 +51,15 @@ def get_exchange(
    ----------

    # exchange related arguments
-    exchange: Exchange(). It could be None or any types that are acceptable by `init_instance_by_config`.
+    exchange: Exchange
+        It could be None or any types that are acceptable by `init_instance_by_config`.
    freq: str
        frequency of data.
    start_time: Union[pd.Timestamp, str]
        closed start time for backtest.
    end_time: Union[pd.Timestamp, str]
        closed end time for backtest.
-    codes: list|str
+    codes: Union[list, str]
        list stock_id list or a string of instruments (i.e. all, csi500, sse50)
    subscribe_fields: list
        subscribe fields.
@@ -151,28 +151,24 @@ def create_account_instance(
        Postion type.
    """
    if isinstance(account, (int, float)):
-        pos_kwargs = {"init_cash": account}
+        init_cash = account
+        position_dict = {}
    elif isinstance(account, dict):
-        init_cash = account["cash"]
-        del account["cash"]
-        pos_kwargs = {
-            "init_cash": init_cash,
-            "position_dict": account,
-        }
+        init_cash = account.pop("cash")
+        position_dict = account
    else:
-        raise ValueError("account must be in (int, float, Position)")
+        raise ValueError("account must be in (int, float, dict)")

-    kwargs = {
-        "init_cash": account,
-        "benchmark_config": {
+    return Account(
+        init_cash=init_cash,
+        position_dict=position_dict,
+        pos_type=pos_type,
+        benchmark_config={
            "benchmark": benchmark,
            "start_time": start_time,
            "end_time": end_time,
        },
-        "pos_type": pos_type,
-    }
-    kwargs.update(pos_kwargs)
-    return Account(**kwargs)
+    )


 def get_strategy_executor(
@@ -181,7 +177,7 @@ def get_strategy_executor(
    strategy: Union[str, dict, object, Path],
    executor: Union[str, dict, object, Path],
    benchmark: str = "SH000300",
-    account: Union[float, int, Position] = 1e9,
+    account: Union[float, int, dict] = 1e9,
    exchange_kwargs: dict = {},
    pos_type: str = "Position",
 ) -> Tuple[BaseStrategy, BaseExecutor]:
@@ -222,7 +218,7 @@ def backtest(
    strategy: Union[str, dict, object, Path],
    executor: Union[str, dict, object, Path],
    benchmark: str = "SH000300",
-    account: Union[float, int, Position] = 1e9,
+    account: Union[float, int, dict] = 1e9,
    exchange_kwargs: dict = {},
    pos_type: str = "Position",
 ) -> Tuple[PortfolioMetrics, Indicator]:
@@ -285,7 +281,7 @@ def collect_data(
    strategy: Union[str, dict, object, Path],
    executor: Union[str, dict, object, Path],
    benchmark: str = "SH000300",
-    account: Union[float, int, Position] = 1e9,
+    account: Union[float, int, dict] = 1e9,
    exchange_kwargs: dict = {},
    pos_type: str = "Position",
    return_value: dict = None,
@@ -339,7 +335,7 @@ def format_decisions(

    cur_freq = decisions[0].strategy.trade_calendar.get_freq()

-    res = (cur_freq, [])
+    res: Tuple[str, list] = (cur_freq, [])
    last_dec_idx = 0
    for i, dec in enumerate(decisions[1:], 1):
        if dec.strategy.trade_calendar.get_freq() == cur_freq:
--- a/qlib/backtest/account.py
+++ b/qlib/backtest/account.py
@@ -3,7 +3,7 @@
 from __future__ import annotations

 import copy
-from typing import Dict, List, Tuple
+from typing import Dict, List, Optional, Tuple, cast

 import pandas as pd

@@ -11,6 +11,7 @@ from qlib.utils import init_instance_by_config

 from .decision import BaseTradeDecision, Order
 from .exchange import Exchange
+from .high_performance_ds import BaseOrderIndicator
 from .position import BasePosition
 from .report import Indicator, PortfolioMetrics

@@ -104,7 +105,7 @@ class Account:

        self._pos_type = pos_type
        self._port_metr_enabled = port_metr_enabled
-        self.benchmark_config = None  # avoid no attribute error
+        self.benchmark_config: dict = {}  # avoid no attribute error
        self.init_vars(init_cash, position_dict, freq, benchmark_config)

    def init_vars(self, init_cash: float, position_dict: dict, freq: str, benchmark_config: dict) -> None:
@@ -124,8 +125,8 @@ class Account:
        self.accum_info = AccumulatedInfo()

        # 2) following variables are not shared between layers
-        self.portfolio_metrics = None
-        self.hist_positions = {}
+        self.portfolio_metrics: Optional[PortfolioMetrics] = None
+        self.hist_positions: Dict[pd.Timestamp, BasePosition] = {}
        self.reset(freq=freq, benchmark_config=benchmark_config)

    def is_port_metr_enabled(self) -> bool:
@@ -171,7 +172,7 @@ class Account:

        self.reset_report(self.freq, self.benchmark_config)

-    def get_hist_positions(self) -> dict:
+    def get_hist_positions(self) -> Dict[pd.Timestamp, BasePosition]:
        return self.hist_positions

    def get_cash(self) -> float:
@@ -230,13 +231,15 @@ class Account:
        """
        # update price for stock in the position and the profit from changed_price
        # NOTE: updating position does not only serve portfolio metrics, it also serve the strategy
+        assert self.current_position is not None
+
        if not self.current_position.skip_update():
            stock_list = self.current_position.get_stock_list()
            for code in stock_list:
                # if suspend, no new price to be updated, profit is 0
                if trade_exchange.check_stock_suspended(code, trade_start_time, trade_end_time):
                    continue
-                bar_close = trade_exchange.get_close(code, trade_start_time, trade_end_time)
+                bar_close = cast(float, trade_exchange.get_close(code, trade_start_time, trade_end_time))
                self.current_position.update_stock_price(stock_id=code, price=bar_close)
            # update holding day count
            # NOTE: updating bar_count does not only serve portfolio metrics, it also serve the strategy
@@ -249,6 +252,8 @@ class Account:
        # for the first trade date, account_value - init_cash
        # self.portfolio_metrics.is_empty() to judge is_first_trade_date
        # get last_account_value, last_total_cost, last_total_turnover
+        assert self.portfolio_metrics is not None
+
        if self.portfolio_metrics.is_empty():
            last_account_value = self.init_cash
            last_total_cost = 0
@@ -299,9 +304,9 @@ class Account:
        trade_exchange: Exchange,
        atomic: bool,
        outer_trade_decision: BaseTradeDecision,
-        trade_info: list = None,
-        inner_order_indicators: List[Dict[str, pd.Series]] = None,
-        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]] = None,
+        trade_info: list = [],
+        inner_order_indicators: List[BaseOrderIndicator] = [],
+        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]] = [],
        indicator_config: dict = {},
    ) -> None:
        """update trade indicators and order indicators in each bar end"""
@@ -335,9 +340,9 @@ class Account:
        trade_exchange: Exchange,
        atomic: bool,
        outer_trade_decision: BaseTradeDecision,
-        trade_info: list = None,
-        inner_order_indicators: List[Dict[str, pd.Series]] = None,
-        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]] = None,
+        trade_info: list = [],
+        inner_order_indicators: List[BaseOrderIndicator] = [],
+        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]] = [],
        indicator_config: dict = {},
    ) -> None:
        """update account at each trading bar step
@@ -398,6 +403,7 @@ class Account:
    def get_portfolio_metrics(self) -> Tuple[pd.DataFrame, dict]:
        """get the history portfolio_metrics and positions instance"""
        if self.is_port_metr_enabled():
+            assert self.portfolio_metrics is not None
            _portfolio_metrics = self.portfolio_metrics.generate_portfolio_metrics_dataframe()
            _positions = self.get_hist_positions()
            return _portfolio_metrics, _positions
--- a/qlib/backtest/backtest.py
+++ b/qlib/backtest/backtest.py
@@ -3,7 +3,7 @@

 from __future__ import annotations

-from typing import TYPE_CHECKING, Generator, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Generator, Optional, Tuple, Union, cast

 import pandas as pd

@@ -36,10 +36,13 @@ def backtest_loop(
    indicator: Indicator
        it computes the trading indicator
    """
-    return_value = {}
+    return_value: dict = {}
    for _decision in collect_data_loop(start_time, end_time, trade_strategy, trade_executor, return_value):
        pass
-    return return_value.get("portfolio_metrics"), return_value.get("indicator")
+
+    portfolio_metrics = cast(PortfolioMetrics, return_value.get("portfolio_metrics"))
+    indicator = cast(Indicator, return_value.get("indicator"))
+    return portfolio_metrics, indicator


 def collect_data_loop(
--- a/qlib/backtest/decision.py
+++ b/qlib/backtest/decision.py
@@ -7,7 +7,7 @@ from abc import abstractmethod
 from enum import IntEnum

 # try to fix circular imports when enabling type hints
-from typing import TYPE_CHECKING, ClassVar, List, Optional, Tuple, Union
+from typing import Generic, List, TYPE_CHECKING, Any, ClassVar, Optional, Tuple, TypeVar, Union, cast

 from qlib.backtest.utils import TradeCalendarManager
 from qlib.data.data import Cal
@@ -24,8 +24,11 @@ import numpy as np
 import pandas as pd


+DecisionType = TypeVar("DecisionType")
+
+
 class OrderDir(IntEnum):
-    # Order  direction
+    # Order direction
    SELL = 0
    BUY = 1

@@ -65,7 +68,7 @@ class Order:
    # - not tradable: the deal_amount == 0 , factor is None
    #    - the stock is suspended and the entire order fails. No cost for this order
    # - dealt or partially dealt: deal_amount >= 0 and factor is not None
-    deal_amount: Optional[float] = None  # `deal_amount` is a non-negative value
+    deal_amount: float = 0.0  # `deal_amount` is a non-negative value
    factor: Optional[float] = None

    # TODO:
@@ -281,7 +284,7 @@ class TradeRangeByTime(TradeRange):
        return max(val_start, start_time), min(val_end, end_time)


-class BaseTradeDecision:
+class BaseTradeDecision(Generic[DecisionType]):
    """
    Trade decisions ara made by strategy and executed by executor

@@ -316,20 +319,21 @@ class BaseTradeDecision:
        """
        self.strategy = strategy
        self.start_time, self.end_time = strategy.trade_calendar.get_step_time()
-        self.total_step = None  # upper strategy has no knowledge about the sub executor before `_init_sub_trading`
-        if isinstance(trade_range, Tuple):
+        # upper strategy has no knowledge about the sub executor before `_init_sub_trading`
+        self.total_step: Optional[int] = None
+        if isinstance(trade_range, tuple):
            # for Tuple[int, int]
            trade_range = IdxTradeRange(*trade_range)
-        self.trade_range: TradeRange = trade_range
+        self.trade_range: Optional[TradeRange] = trade_range

-    def get_decision(self) -> List[object]:
+    def get_decision(self) -> List[DecisionType]:
        """
        get the **concrete decision**  (e.g. execution orders)
        This will be called by the inner strategy

        Returns
        -------
-        List[object]:
+        List[DecisionType:
            The decision result. Typically it is some orders
            Example:
                []:
@@ -363,13 +367,13 @@ class BaseTradeDecision:
        # purpose 2)
        return self.strategy.update_trade_decision(self, trade_calendar)

-    def _get_range_limit(self, **kwargs) -> Tuple[int, int]:
+    def _get_range_limit(self, **kwargs: Any) -> Tuple[int, int]:
        if self.trade_range is not None:
-            return self.trade_range(trade_calendar=kwargs.get("inner_calendar"))
+            return self.trade_range(trade_calendar=cast(TradeCalendarManager, kwargs.get("inner_calendar")))
        else:
            raise NotImplementedError("The decision didn't provide an index range")

-    def get_range_limit(self, **kwargs) -> Tuple[int, int]:
+    def get_range_limit(self, **kwargs: Any) -> Tuple[int, int]:
        """
        return the expected step range for limiting the decision execution time
        Both left and right are **closed**
@@ -421,6 +425,7 @@ class BaseTradeDecision:
        if getattr(self, "total_step", None) is not None:
            # if `self.update` is called.
            # Then the _start_idx, _end_idx should be clipped
+            assert self.total_step is not None
            if _start_idx < 0 or _end_idx >= self.total_step:
                logger = get_module_logger("decision")
                logger.warning(
@@ -516,7 +521,7 @@ class BaseTradeDecision:
            inner_trade_decision.trade_range = self.trade_range


-class EmptyTradeDecision(BaseTradeDecision):
+class EmptyTradeDecision(BaseTradeDecision[object]):
    def get_decision(self) -> List[object]:
        return []

@@ -524,23 +529,24 @@ class EmptyTradeDecision(BaseTradeDecision):
        return True


-class TradeDecisionWO(BaseTradeDecision):
+class TradeDecisionWO(BaseTradeDecision[Order]):
    """
    Trade Decision (W)ith (O)rder.
    Besides, the time_range is also included.
    """

-    def __init__(self, order_list: List[Order], strategy: BaseStrategy, trade_range: Tuple[int, int] = None):
+    def __init__(self, order_list: List[object], strategy: BaseStrategy, trade_range: Tuple[int, int] = None) -> None:
        super().__init__(strategy, trade_range=trade_range)
-        self.order_list = order_list
+        self.order_list = cast(List[Order], order_list)
        start, end = strategy.trade_calendar.get_step_time()
        for o in order_list:
+            assert isinstance(o, Order)
            if o.start_time is None:
                o.start_time = start
            if o.end_time is None:
                o.end_time = end

-    def get_decision(self) -> List[object]:
+    def get_decision(self) -> List[Order]:
        return self.order_list

    def __repr__(self) -> str:
--- a/qlib/backtest/exchange.py
+++ b/qlib/backtest/exchange.py
@@ -3,7 +3,7 @@
 from __future__ import annotations

 from collections import defaultdict
-from typing import TYPE_CHECKING, List, Optional, Tuple, Type, Union
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Type, Union, cast

 from ..utils.index_data import IndexData

@@ -42,7 +42,7 @@ class Exchange:
        impact_cost: float = 0.0,
        extra_quote: pd.DataFrame = None,
        quote_cls: Type[BaseQuote] = NumpyQuote,
-        **kwargs,
+        **kwargs: Any,
    ) -> None:
        """__init__
        :param freq:             frequency of data
@@ -141,7 +141,7 @@ class Exchange:
        if limit_threshold is None:
            if C.region == REG_CN:
                self.logger.warning(f"limit_threshold not set. The stocks hit the limit may be bought/sold")
-        elif self.limit_type == self.LT_FLT and abs(limit_threshold) > 0.1:
+        elif self.limit_type == self.LT_FLT and abs(cast(float, limit_threshold)) > 0.1:
            if C.region == REG_CN:
                self.logger.warning(f"limit_threshold may not be set to a reasonable value")

@@ -150,7 +150,7 @@ class Exchange:
                deal_price = "$" + deal_price
            self.buy_price = self.sell_price = deal_price
        elif isinstance(deal_price, (tuple, list)):
-            self.buy_price, self.sell_price = deal_price
+            self.buy_price, self.sell_price = cast(Tuple[str, str], deal_price)
        else:
            raise NotImplementedError(f"This type of input is not supported")

@@ -167,10 +167,10 @@ class Exchange:

        necessary_fields = {self.buy_price, self.sell_price, "$close", "$change", "$factor", "$volume"}
        if self.limit_type == self.LT_TP_EXP:
+            assert isinstance(limit_threshold, tuple)
            for exp in limit_threshold:
                necessary_fields.add(exp)
-        all_fields = necessary_fields | set(vol_lt_fields)
-        all_fields = list(all_fields | set(subscribe_fields))
+        all_fields = list(necessary_fields | set(vol_lt_fields) | set(subscribe_fields))

        self.all_fields = all_fields

@@ -249,9 +249,9 @@ class Exchange:
    LT_FLT = "float"  # float
    LT_NONE = "none"  # none

-    def _get_limit_type(self, limit_threshold: Union[Tuple, float, None]) -> str:
+    def _get_limit_type(self, limit_threshold: Union[tuple, float, None]) -> str:
        """get limit type"""
-        if isinstance(limit_threshold, Tuple):
+        if isinstance(limit_threshold, tuple):
            return self.LT_TP_EXP
        elif isinstance(limit_threshold, float):
            return self.LT_FLT
@@ -268,14 +268,16 @@ class Exchange:
            self.quote_df["limit_sell"] = False
        elif limit_type == self.LT_TP_EXP:
            # set limit
+            limit_threshold = cast(tuple, limit_threshold)
            self.quote_df["limit_buy"] = self.quote_df[limit_threshold[0]]
            self.quote_df["limit_sell"] = self.quote_df[limit_threshold[1]]
        elif limit_type == self.LT_FLT:
+            limit_threshold = cast(float, limit_threshold)
            self.quote_df["limit_buy"] = self.quote_df["$change"].ge(limit_threshold)
            self.quote_df["limit_sell"] = self.quote_df["$change"].le(-limit_threshold)  # pylint: disable=E1130

    @staticmethod
-    def _get_vol_limit(volume_threshold: Union[tuple, dict]) -> Tuple[Optional[list], Optional[list], set]:
+    def _get_vol_limit(volume_threshold: Union[tuple, dict, None]) -> Tuple[Optional[list], Optional[list], set]:
        """
        preprocess the volume limit.
        get the fields need to get from qlib.
@@ -340,11 +342,11 @@ class Exchange:
        if direction is None:
            buy_limit = self.quote.get_data(stock_id, start_time, end_time, field="limit_buy", method="all")
            sell_limit = self.quote.get_data(stock_id, start_time, end_time, field="limit_sell", method="all")
-            return buy_limit or sell_limit
+            return bool(buy_limit or sell_limit)
        elif direction == Order.BUY:
-            return self.quote.get_data(stock_id, start_time, end_time, field="limit_buy", method="all")
+            return cast(bool, self.quote.get_data(stock_id, start_time, end_time, field="limit_buy", method="all"))
        elif direction == Order.SELL:
-            return self.quote.get_data(stock_id, start_time, end_time, field="limit_sell", method="all")
+            return cast(bool, self.quote.get_data(stock_id, start_time, end_time, field="limit_sell", method="all"))
        else:
            raise ValueError(f"direction {direction} is not supported!")

@@ -382,7 +384,7 @@ class Exchange:
        order: Order,
        trade_account: Account = None,
        position: BasePosition = None,
-        dealt_order_amount: defaultdict = defaultdict(float),
+        dealt_order_amount: Dict[str, float] = defaultdict(float),
    ) -> Tuple[float, float, float]:
        """
        Deal order when the actual transaction
@@ -426,9 +428,10 @@ class Exchange:
        stock_id: str,
        start_time: pd.Timestamp,
        end_time: pd.Timestamp,
+        field: str,
        method: str = "ts_data_last",
    ) -> Union[None, int, float, bool, IndexData]:
-        return self.quote.get_data(stock_id, start_time, end_time, method=method)  # TODO: missing `field`?
+        return self.quote.get_data(stock_id, start_time, end_time, field=field, method=method)

    def get_close(
        self,
@@ -444,10 +447,10 @@ class Exchange:
        stock_id: str,
        start_time: pd.Timestamp,
        end_time: pd.Timestamp,
-        method: str = "sum",
+        method: Optional[str] = "sum",
    ) -> float:
        """get the total deal volume of stock with `stock_id` between the time interval [start_time, end_time)"""
-        return self.quote.get_data(stock_id, start_time, end_time, field="$volume", method=method)
+        return cast(float, self.quote.get_data(stock_id, start_time, end_time, field="$volume", method=method))

    def get_deal_price(
        self,
@@ -455,7 +458,7 @@ class Exchange:
        start_time: pd.Timestamp,
        end_time: pd.Timestamp,
        direction: OrderDir,
-        method: str = "ts_data_last",
+        method: Optional[str] = "ts_data_last",
    ) -> float:
        if direction == OrderDir.SELL:
            pstr = self.sell_price
@@ -469,7 +472,7 @@ class Exchange:
            self.logger.warning(f"(stock_id:{stock_id}, trade_time:{(start_time, end_time)}, {pstr}): {deal_price}!!!")
            self.logger.warning(f"setting deal_price to close price")
            deal_price = self.get_close(stock_id, start_time, end_time, method)
-        return deal_price
+        return cast(float, deal_price)

    def get_factor(
        self,
@@ -544,7 +547,7 @@ class Exchange:
                )
        return amount_dict

-    def get_real_deal_amount(self, current_amount: float, target_amount: float, factor: float) -> float:
+    def get_real_deal_amount(self, current_amount: float, target_amount: float, factor: float = None) -> float:
        """
        Calculate the real adjust deal amount when considering the trading unit
        :param current_amount:
@@ -572,7 +575,7 @@ class Exchange:
        current_position: dict,
        start_time: pd.Timestamp,
        end_time: pd.Timestamp,
-    ) -> list:
+    ) -> List[Order]:
        """
        Note: some future information is used in this function
        Parameter:
@@ -681,6 +684,7 @@ class Exchange:
                factor = self.get_factor(stock_id=stock_id, start_time=start_time, end_time=end_time)
            else:
                raise ValueError(f"`factor` and (`stock_id`, `start_time`, `end_time`) can't both be None")
+        assert factor is not None
        return factor

    def get_amount_of_trade_unit(
@@ -718,12 +722,12 @@ class Exchange:

    def round_amount_by_trade_unit(
        self,
-        deal_amount,
+        deal_amount: float,
        factor: float = None,
        stock_id: str = None,
-        start_time=None,
-        end_time=None,
-    ):
+        start_time: pd.Timestamp = None,
+        end_time: pd.Timestamp = None,
+    ) -> float:
        """Parameter
        Please refer to the docs of get_amount_of_trade_unit
        deal_amount : float, adjusted amount
@@ -741,7 +745,7 @@ class Exchange:
            return (deal_amount * factor + 0.1) // self.trade_unit * self.trade_unit / factor
        return deal_amount

-    def _clip_amount_by_volume(self, order: Order, dealt_order_amount: dict) -> int:
+    def _clip_amount_by_volume(self, order: Order, dealt_order_amount: dict) -> Optional[float]:
        """parse the capacity limit string and return the actual amount of orders that can be executed.
        NOTE:
            this function will change the order.deal_amount **inplace**
@@ -753,15 +757,12 @@ class Exchange:
        dealt_order_amount : dict
            :param dealt_order_amount: the dealt order amount dict with the format of {stock_id: float}
        """
-        if order.direction == Order.BUY:
-            vol_limit = self.buy_vol_limit
-        elif order.direction == Order.SELL:
-            vol_limit = self.sell_vol_limit
+        vol_limit = self.buy_vol_limit if order.direction == Order.BUY else self.sell_vol_limit

        if vol_limit is None:
            return order.deal_amount

-        vol_limit_num = []
+        vol_limit_num: List[float] = []
        for limit in vol_limit:
            assert isinstance(limit, tuple)
            if limit[0] == "current":
@@ -772,7 +773,7 @@ class Exchange:
                    field=limit[1],
                    method="sum",
                )
-                vol_limit_num.append(limit_value)
+                vol_limit_num.append(cast(float, limit_value))
            elif limit[0] == "cum":
                limit_value = self.quote.get_data(
                    order.stock_id,
@@ -790,12 +791,14 @@ class Exchange:
        if vol_limit_min < orig_deal_amount:
            self.logger.debug(f"Order clipped due to volume limitation: {order}, {list(zip(vol_limit_num, vol_limit))}")

-    def _get_buy_amount_by_cash_limit(self, trade_price, cash, cost_ratio):
+        return None
+
+    def _get_buy_amount_by_cash_limit(self, trade_price: float, cash: float, cost_ratio: float) -> float:
        """return the real order amount after cash limit for buying.
        Parameters
        ----------
        trade_price : float
-        position : cash
+        cash : float
        cost_ratio : float

        Return
@@ -803,7 +806,7 @@ class Exchange:
        float
            the real order amount after cash limit for buying.
        """
-        max_trade_amount = 0
+        max_trade_amount = 0.0
        if cash >= self.min_cost:
            # critical_price means the stock transaction price when the service fee is equal to min_cost.
            critical_price = self.min_cost / cost_ratio + self.min_cost
@@ -897,7 +900,7 @@ class Exchange:
                order.deal_amount = self.round_amount_by_trade_unit(order.deal_amount, order.factor)

        else:
-            raise NotImplementedError("order type {} error".format(order.type))
+            raise NotImplementedError("order direction {} error".format(order.direction))

        trade_val = order.deal_amount * trade_price
        trade_cost = max(trade_val * cost_ratio, self.min_cost)
--- a/qlib/backtest/executor.py
+++ b/qlib/backtest/executor.py
@@ -4,7 +4,7 @@ import copy
 from abc import abstractmethod
 from collections import defaultdict
 from types import GeneratorType
-from typing import Generator, List, Optional, Tuple, Union
+from typing import Any, Dict, Generator, List, Tuple, Union, cast

 import pandas as pd

@@ -16,13 +16,7 @@ from ..strategy.base import BaseStrategy
 from ..utils import init_instance_by_config
 from .decision import BaseTradeDecision, Order
 from .exchange import Exchange
-from .utils import (
-    BaseInfrastructure,
-    CommonInfrastructure,
-    LevelInfrastructure,
-    TradeCalendarManager,
-    get_start_end_idx,
-)
+from .utils import CommonInfrastructure, LevelInfrastructure, TradeCalendarManager, get_start_end_idx


 class BaseExecutor:
@@ -39,8 +33,8 @@ class BaseExecutor:
        track_data: bool = False,
        trade_exchange: Exchange = None,
        common_infra: CommonInfrastructure = None,
-        settle_type=BasePosition.ST_NO,  # TODO: add typehint
-        **kwargs,
+        settle_type: str = BasePosition.ST_NO,
+        **kwargs: Any,
    ) -> None:
        """
        Parameters
@@ -127,10 +121,10 @@ class BaseExecutor:
            get_module_logger("BaseExecutor").warning(f"`common_infra` is not set for {self}")

        # record deal order amount in one day
-        self.dealt_order_amount = defaultdict(float)
+        self.dealt_order_amount: Dict[str, float] = defaultdict(float)
        self.deal_day = None

-    def reset_common_infra(self, common_infra: BaseInfrastructure, copy_trade_account: bool = False) -> None:
+    def reset_common_infra(self, common_infra: CommonInfrastructure, copy_trade_account: bool = False) -> None:
        """
        reset infrastructure for trading
            - reset trade_account
@@ -141,14 +135,15 @@ class BaseExecutor:
            self.common_infra.update(common_infra)

        if common_infra.has("trade_account"):
-            if copy_trade_account:
-                # NOTE: there is a trick in the code.
-                # shallow copy is used instead of deepcopy.
-                # 1. So positions are shared
-                # 2. Others are not shared, so each level has it own metrics (portfolio and trading metrics)
-                self.trade_account: Account = copy.copy(common_infra.get("trade_account"))
-            else:
-                self.trade_account: Account = common_infra.get("trade_account")
+            # NOTE: there is a trick in the code.
+            # shallow copy is used instead of deepcopy.
+            # 1. So positions are shared
+            # 2. Others are not shared, so each level has it own metrics (portfolio and trading metrics)
+            self.trade_account: Account = (
+                copy.copy(common_infra.get("trade_account"))
+                if copy_trade_account
+                else common_infra.get("trade_account")
+            )
            self.trade_account.reset(freq=self.time_per_step, port_metr_enabled=self.generate_portfolio_metrics)

    @property
@@ -164,7 +159,7 @@ class BaseExecutor:
        """
        return self.level_infra.get("trade_calendar")

-    def reset(self, common_infra: CommonInfrastructure = None, **kwargs) -> None:
+    def reset(self, common_infra: CommonInfrastructure = None, **kwargs: Any) -> None:
        """
        - reset `start_time` and `end_time`, used in trade calendar
        - reset `common_infra`, used to reset `trade_account`, `trade_exchange`, .etc
@@ -200,20 +195,17 @@ class BaseExecutor:
        execute_result : List[object]
            the executed result for trade decision
        """
-        return_value = {}
+        return_value: dict = {}
        for _decision in self.collect_data(trade_decision, return_value=return_value, level=level):
            pass
-        return return_value.get("execute_result")
+        return cast(list, return_value.get("execute_result"))

    @abstractmethod
    def _collect_data(
        self,
        trade_decision: BaseTradeDecision,
        level: int = 0,
-    ) -> Union[
-        Generator[BaseTradeDecision, Optional[BaseTradeDecision], Tuple[List[object], dict]],
-        Tuple[List[object], dict],
-    ]:
+    ) -> Union[Generator[Any, Any, Tuple[List[object], dict]], Tuple[List[object], dict]]:
        """
        Please refer to the doc of collect_data
        The only difference between `_collect_data` and `collect_data` is that some common steps are moved into
@@ -235,7 +227,7 @@ class BaseExecutor:
        trade_decision: BaseTradeDecision,
        return_value: dict = None,
        level: int = 0,
-    ) -> Generator[BaseTradeDecision, Optional[BaseTradeDecision], List[object]]:
+    ) -> Generator[Any, Any, List[object]]:
        """Generator for collecting the trade decision data for rl training

        his function will make a step forward
@@ -332,7 +324,7 @@ class NestedExecutor(BaseExecutor):
        skip_empty_decision: bool = True,
        align_range_limit: bool = True,
        common_infra: CommonInfrastructure = None,
-        **kwargs,
+        **kwargs: Any,
    ) -> None:
        """
        Parameters
@@ -411,7 +403,7 @@ class NestedExecutor(BaseExecutor):
        self,
        trade_decision: BaseTradeDecision,
        level: int = 0,
-    ) -> Generator[BaseTradeDecision, Optional[BaseTradeDecision], Tuple[List[object], dict]]:
+    ) -> Generator[Any, Any, Tuple[List[object], dict]]:
        execute_result = []
        inner_order_indicators = []
        decision_list = []
@@ -493,7 +485,7 @@ class NestedExecutor(BaseExecutor):
            the execution result of inner task
        """

-    def get_all_executors(self) -> List[object]:
+    def get_all_executors(self) -> List[BaseExecutor]:
        """get all executors, including self and inner_executor.get_all_executors()"""
        return [self, *self.inner_executor.get_all_executors()]

@@ -536,7 +528,7 @@ class SimulatorExecutor(BaseExecutor):
        track_data: bool = False,
        common_infra: CommonInfrastructure = None,
        trade_type: str = TT_SERIAL,
-        **kwargs,
+        **kwargs: Any,
    ) -> None:
        """
        Parameters
@@ -598,7 +590,7 @@ class SimulatorExecutor(BaseExecutor):

    def _collect_data(self, trade_decision: BaseTradeDecision, level: int = 0) -> Tuple[List[object], dict]:
        trade_start_time, _ = self.trade_calendar.get_step_time()
-        execute_result = []
+        execute_result: list = []

        for order in self._get_order_iterator(trade_decision):
            # execute the order.
--- a/qlib/backtest/high_performance_ds.py
+++ b/qlib/backtest/high_performance_ds.py
@@ -1,11 +1,13 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT License.

+from __future__ import annotations
+
 import inspect
 import logging
 from collections import OrderedDict
 from functools import lru_cache
-from typing import Callable, Dict, Iterable, List, Text, Union
+from typing import Any, Callable, Dict, Iterable, List, Optional, Text, Union, cast

 import numpy as np
 import pandas as pd
@@ -19,7 +21,7 @@ from ..utils.time import Freq, is_single_value


 class BaseQuote:
-    def __init__(self, quote_df: pd.DataFrame, freq):
+    def __init__(self, quote_df: pd.DataFrame, freq: str) -> None:
        self.logger = get_module_logger("online operator", level=logging.INFO)

    def get_all_stock(self) -> Iterable:
@@ -39,7 +41,7 @@ class BaseQuote:
        start_time: Union[pd.Timestamp, str],
        end_time: Union[pd.Timestamp, str],
        field: Union[str],
-        method: Union[str, None] = None,
+        method: Optional[str] = None,
    ) -> Union[None, int, float, bool, IndexData]:
        """get the specific field of stock data during start time and end_time,
           and apply method to the data.
@@ -99,7 +101,7 @@ class BaseQuote:


 class PandasQuote(BaseQuote):
-    def __init__(self, quote_df: pd.DataFrame, freq):
+    def __init__(self, quote_df: pd.DataFrame, freq: str) -> None:
        super().__init__(quote_df=quote_df, freq=freq)
        quote_dict = {}
        for stock_id, stock_val in quote_df.groupby(level="instrument"):
@@ -124,7 +126,7 @@ class PandasQuote(BaseQuote):


 class NumpyQuote(BaseQuote):
-    def __init__(self, quote_df: pd.DataFrame, freq, region="cn"):
+    def __init__(self, quote_df: pd.DataFrame, freq: str, region: str = "cn") -> None:
        """NumpyQuote

        Parameters
@@ -178,7 +180,8 @@ class NumpyQuote(BaseQuote):
                data = self._agg_data(data, method)
            return data

-    def _agg_data(self, data: IndexData, method):
+    @staticmethod
+    def _agg_data(data: IndexData, method: str) -> Union[IndexData, np.ndarray, None]:
        """Agg data by specific method."""
        # FIXME: why not call the method of data directly?
        if method == "sum":
@@ -224,31 +227,31 @@ class BaseSingleMetric:
        """
        raise NotImplementedError(f"Please implement the `__init__` method")

-    def __add__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __add__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__add__` method")

-    def __radd__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __radd__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        return self + other

-    def __sub__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __sub__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__sub__` method")

-    def __rsub__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __rsub__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__rsub__` method")

-    def __mul__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __mul__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__mul__` method")

-    def __truediv__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __truediv__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__truediv__` method")

-    def __eq__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __eq__(self, other: object) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__eq__` method")

-    def __gt__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __gt__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__gt__` method")

-    def __lt__(self, other: Union["BaseSingleMetric", int, float]) -> "BaseSingleMetric":
+    def __lt__(self, other: Union[BaseSingleMetric, int, float]) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `__lt__` method")

    def __len__(self) -> int:
@@ -265,7 +268,7 @@ class BaseSingleMetric:

        raise NotImplementedError(f"Please implement the `count` method")

-    def abs(self) -> "BaseSingleMetric":
+    def abs(self) -> BaseSingleMetric:
        raise NotImplementedError(f"Please implement the `abs` method")

    @property
@@ -274,18 +277,18 @@ class BaseSingleMetric:

        raise NotImplementedError(f"Please implement the `empty` method")

-    def add(self, other: "BaseSingleMetric", fill_value: float = None) -> "BaseSingleMetric":
+    def add(self, other: BaseSingleMetric, fill_value: float = None) -> BaseSingleMetric:
        """Replace np.NaN with fill_value in two metrics and add them."""

        raise NotImplementedError(f"Please implement the `add` method")

-    def replace(self, replace_dict: dict) -> "BaseSingleMetric":
+    def replace(self, replace_dict: dict) -> BaseSingleMetric:
        """Replace the value of metric according to replace_dict."""

        raise NotImplementedError(f"Please implement the `replace` method")

-    def apply(self, func: dict) -> "BaseSingleMetric":
-        """Replace the value of metric with func(metric).
+    def apply(self, func: Callable) -> BaseSingleMetric:
+        """Replace the value of metric with func (metric).
        Currently, the func is only qlib/backtest/order/Order.parse_dir.
        """

@@ -304,11 +307,11 @@ class BaseOrderIndicator:
        to inherit the BaseSingleMetric.
    """

-    def __init__(self, data):
-        self.data = data
+    def __init__(self):
+        self.data = {}  # will be created in the subclass
        self.logger = get_module_logger("online operator")

-    def assign(self, col: str, metric: Union[dict, pd.Series]):
+    def assign(self, col: str, metric: Union[dict, pd.Series]) -> None:
        """assign one metric.

        Parameters
@@ -328,7 +331,7 @@ class BaseOrderIndicator:

        raise NotImplementedError(f"Please implement the 'assign' method")

-    def transfer(self, func: Callable, new_col: str = None) -> Union[None, BaseSingleMetric]:
+    def transfer(self, func: Callable, new_col: str = None) -> Optional[BaseSingleMetric]:
        """compute new metric with existing metrics.

        Parameters
@@ -352,6 +355,7 @@ class BaseOrderIndicator:
        tmp_metric = func(**func_kwargs)
        if new_col is not None:
            self.data[new_col] = tmp_metric
+            return None
        else:
            return tmp_metric

@@ -372,7 +376,7 @@ class BaseOrderIndicator:

        raise NotImplementedError(f"Please implement the 'get_metric_series' method")

-    def get_index_data(self, metric) -> SingleData:
+    def get_index_data(self, metric: str) -> SingleData:
        """get one metric with the format of SingleData

        Parameters
@@ -389,7 +393,12 @@ class BaseOrderIndicator:
        raise NotImplementedError(f"Please implement the 'get_index_data' method")

    @staticmethod
-    def sum_all_indicators(order_indicator, indicators: list, metrics: Union[str, List[str]], fill_value: float = None):
+    def sum_all_indicators(
+        order_indicator: BaseOrderIndicator,
+        indicators: List[BaseOrderIndicator],
+        metrics: Union[str, List[str]],
+        fill_value: float = 0,
+    ) -> None:
        """sum indicators with the same metrics.
        and assign to the order_indicator(BaseOrderIndicator).
        NOTE: indicators could be a empty list when orders in lower level all fail.
@@ -527,16 +536,17 @@ class PandasSingleMetric(SingleMetric):
    def index(self):
        return list(self.metric.index)

-    def add(self, other, fill_value=None):
+    def add(self, other: BaseSingleMetric, fill_value: float = None) -> PandasSingleMetric:
+        other = cast(PandasSingleMetric, other)
        return self.__class__(self.metric.add(other.metric, fill_value=fill_value))

-    def replace(self, replace_dict: dict):
+    def replace(self, replace_dict: dict) -> PandasSingleMetric:
        return self.__class__(self.metric.replace(replace_dict))

-    def apply(self, func: Callable):
+    def apply(self, func: Callable) -> PandasSingleMetric:
        return self.__class__(self.metric.apply(func))

-    def reindex(self, index, fill_value):
+    def reindex(self, index: Any, fill_value: float) -> PandasSingleMetric:
        return self.__class__(self.metric.reindex(index, fill_value=fill_value))

    def __repr__(self):
@@ -550,13 +560,14 @@ class PandasOrderIndicator(BaseOrderIndicator):
    Str is the name of metric.
    """

-    def __init__(self):
+    def __init__(self) -> None:
+        super(PandasOrderIndicator, self).__init__()
        self.data: Dict[str, PandasSingleMetric] = OrderedDict()

-    def assign(self, col: str, metric: Union[dict, pd.Series]):
+    def assign(self, col: str, metric: Union[dict, pd.Series]) -> None:
        self.data[col] = PandasSingleMetric(metric)

-    def get_index_data(self, metric):
+    def get_index_data(self, metric: str) -> SingleData:
        if metric in self.data:
            return idd.SingleData(self.data[metric].metric)
        else:
@@ -572,7 +583,12 @@ class PandasOrderIndicator(BaseOrderIndicator):
        return {k: v.metric for k, v in self.data.items()}

    @staticmethod
-    def sum_all_indicators(order_indicator, indicators: list, metrics: Union[str, List[str]], fill_value=0):
+    def sum_all_indicators(
+        order_indicator: BaseOrderIndicator,
+        indicators: List[BaseOrderIndicator],
+        metrics: Union[str, List[str]],
+        fill_value: float = 0,
+    ) -> None:
        if isinstance(metrics, str):
            metrics = [metrics]
        for metric in metrics:
@@ -592,13 +608,14 @@ class NumpyOrderIndicator(BaseOrderIndicator):
    Str is the name of metric.
    """

-    def __init__(self):
+    def __init__(self) -> None:
+        super(NumpyOrderIndicator, self).__init__()
        self.data: Dict[str, SingleData] = OrderedDict()

-    def assign(self, col: str, metric: dict):
+    def assign(self, col: str, metric: dict) -> None:
        self.data[col] = idd.SingleData(metric)

-    def get_index_data(self, metric):
+    def get_index_data(self, metric: str) -> SingleData:
        if metric in self.data:
            return self.data[metric]
        else:
@@ -614,14 +631,18 @@ class NumpyOrderIndicator(BaseOrderIndicator):
        return tmp_metric_dict

    @staticmethod
-    def sum_all_indicators(order_indicator, indicators: list, metrics: Union[str, List[str]], fill_value=0):
+    def sum_all_indicators(
+        order_indicator: BaseOrderIndicator,
+        indicators: List[BaseOrderIndicator],
+        metrics: Union[str, List[str]],
+        fill_value: float = 0,
+    ) -> None:
        # get all index(stock_id)
-        stocks = set()
+        stock_set: set = set()
        for indicator in indicators:
            # set(np.ndarray.tolist()) is faster than set(np.ndarray)
-            stocks = stocks | set(indicator.data[metrics[0]].index.tolist())
-        stocks = list(stocks)
-        stocks.sort()
+            stock_set = stock_set | set(indicator.data[metrics[0]].index.tolist())
+        stocks = sorted(list(stock_set))

        # add metric by index
        if isinstance(metrics, str):
--- a/qlib/backtest/position.py
+++ b/qlib/backtest/position.py
@@ -3,7 +3,7 @@


 from datetime import timedelta
-from typing import Dict, List, Union
+from typing import Any, Dict, List, Union

 import numpy as np
 import pandas as pd
@@ -18,9 +18,9 @@ class BasePosition:
    Please refer to the `Position` class for the position
    """

-    def __init__(self, *args, cash: float = 0.0, **kwargs) -> None:
+    def __init__(self, *args: Any, cash: float = 0.0, **kwargs: Any) -> None:
        self._settle_type = self.ST_NO
-        self.position = {}
+        self.position: dict = {}

    def fill_stock_value(self, start_time: Union[str, pd.Timestamp], freq: str, last_days: int = 30) -> None:
        pass
@@ -96,13 +96,13 @@ class BasePosition:
    def calculate_value(self) -> float:
        raise NotImplementedError(f"Please implement the `calculate_value` method")

-    def get_stock_list(self) -> List:
+    def get_stock_list(self) -> List[str]:
        """
        Get the list of stocks in the position.
        """
        raise NotImplementedError(f"Please implement the `get_stock_list` method")

-    def get_stock_price(self, code) -> float:
+    def get_stock_price(self, code: str) -> float:
        """
        get the latest price of the stock

@@ -113,7 +113,7 @@ class BasePosition:
        """
        raise NotImplementedError(f"Please implement the `get_stock_price` method")

-    def get_stock_amount(self, code) -> float:
+    def get_stock_amount(self, code: str) -> float:
        """
        get the amount of the stock

@@ -144,7 +144,7 @@ class BasePosition:
        """
        raise NotImplementedError(f"Please implement the `get_cash` method")

-    def get_stock_amount_dict(self) -> Dict:
+    def get_stock_amount_dict(self) -> dict:
        """
        generate stock amount dict {stock_id : amount of stock}

@@ -155,7 +155,7 @@ class BasePosition:
        """
        raise NotImplementedError(f"Please implement the `get_stock_amount_dict` method")

-    def get_stock_weight_dict(self, only_stock: bool = False) -> Dict:
+    def get_stock_weight_dict(self, only_stock: bool = False) -> dict:
        """
        generate stock weight dict {stock_id : value weight of stock in the position}
        it is meaningful in the beginning or the end of each trade step
@@ -174,7 +174,7 @@ class BasePosition:
        """
        raise NotImplementedError(f"Please implement the `get_stock_weight_dict` method")

-    def add_count_all(self, bar) -> None:
+    def add_count_all(self, bar: str) -> None:
        """
        Will be called at the end of each bar on each level

@@ -195,7 +195,7 @@ class BasePosition:
        raise NotImplementedError(f"Please implement the `add_count_all` method")

    ST_CASH = "cash"
-    ST_NO = None
+    ST_NO = "None"  # String is more typehint friendly than None

    def settle_start(self, settle_type: str) -> None:
        """
@@ -220,10 +220,10 @@ class BasePosition:
        """
        raise NotImplementedError(f"Please implement the `settle_commit` method")

-    def __str__(self):
+    def __str__(self) -> str:
        return self.__dict__.__str__()

-    def __repr__(self):
+    def __repr__(self) -> str:
        return self.__dict__.__repr__()


@@ -532,7 +532,7 @@ class InfPosition(BasePosition):
    def calculate_value(self) -> float:
        raise NotImplementedError(f"InfPosition doesn't support calculating value")

-    def get_stock_list(self) -> list:
+    def get_stock_list(self) -> List[str]:
        raise NotImplementedError(f"InfPosition doesn't support stock list position")

    def get_stock_price(self, code: str) -> float:
@@ -545,10 +545,10 @@ class InfPosition(BasePosition):
    def get_cash(self, include_settle: bool = False) -> float:
        return np.inf

-    def get_stock_amount_dict(self) -> Dict:
+    def get_stock_amount_dict(self) -> dict:
        raise NotImplementedError(f"InfPosition doesn't support get_stock_amount_dict")

-    def get_stock_weight_dict(self, only_stock: bool = False) -> Dict:
+    def get_stock_weight_dict(self, only_stock: bool = False) -> dict:
        raise NotImplementedError(f"InfPosition doesn't support get_stock_weight_dict")

    def add_count_all(self, bar: str) -> None:
--- a/qlib/backtest/report.py
+++ b/qlib/backtest/report.py
@@ -4,7 +4,7 @@

 import pathlib
 from collections import OrderedDict
-from typing import Dict, List, Tuple, Union
+from typing import Any, Dict, List, Optional, Text, Tuple, Type, Union, cast

 import numpy as np
 import pandas as pd
@@ -15,7 +15,7 @@ from qlib.backtest.exchange import Exchange

 from ..tests.config import CSI300_BENCH
 from ..utils.resam import get_higher_eq_freq_feature, resam_ts_data
-from .high_performance_ds import BaseOrderIndicator, NumpyOrderIndicator, SingleMetric
+from .high_performance_ds import BaseOrderIndicator, BaseSingleMetric, NumpyOrderIndicator


 class PortfolioMetrics:
@@ -38,7 +38,7 @@ class PortfolioMetrics:
        update report
    """

-    def __init__(self, freq: str = "day", benchmark_config: dict = {}):
+    def __init__(self, freq: str = "day", benchmark_config: dict = {}) -> None:
        """
        Parameters
        ----------
@@ -49,13 +49,17 @@ class PortfolioMetrics:
            - benchmark : Union[str, list, pd.Series]
                - If `benchmark` is pd.Series, `index` is trading date; the value T is the change from T-1 to T.
                    example:
-                        print(D.features(D.instruments('csi500'), ['$close/Ref($close, 1)-1'])['$close/Ref($close, 1)-1'].head())
+                        print(
+                            D.features(D.instruments('csi500'),
+                            ['$close/Ref($close, 1)-1'])['$close/Ref($close, 1)-1'].head()
+                        )
                            2017-01-04    0.011693
                            2017-01-05    0.000721
                            2017-01-06   -0.004322
                            2017-01-09    0.006874
                            2017-01-10   -0.003350
-                - If `benchmark` is list, will use the daily average change of the stock pool in the list as the 'bench'.
+                - If `benchmark` is list, will use the daily average change of the stock pool in the list as the
+                    'bench'.
                - If `benchmark` is str, will use the daily change as the 'bench'.
                benchmark code, default is SH000300 CSI300
            - start_time : Union[str, pd.Timestamp], optional
@@ -70,25 +74,26 @@ class PortfolioMetrics:
        self.init_vars()
        self.init_bench(freq=freq, benchmark_config=benchmark_config)

-    def init_vars(self):
-        self.accounts = OrderedDict()  # account position value for each trade time
-        self.returns = OrderedDict()  # daily return rate for each trade time
-        self.total_turnovers = OrderedDict()  # total turnover for each trade time
-        self.turnovers = OrderedDict()  # turnover for each trade time
-        self.total_costs = OrderedDict()  # total trade cost for each trade time
-        self.costs = OrderedDict()  # trade cost rate for each trade time
-        self.values = OrderedDict()  # value for each trade time
-        self.cashes = OrderedDict()
-        self.benches = OrderedDict()
-        self.latest_pm_time = None  # pd.TimeStamp
+    def init_vars(self) -> None:
+        self.accounts: dict = OrderedDict()  # account position value for each trade time
+        self.returns: dict = OrderedDict()  # daily return rate for each trade time
+        self.total_turnovers: dict = OrderedDict()  # total turnover for each trade time
+        self.turnovers: dict = OrderedDict()  # turnover for each trade time
+        self.total_costs: dict = OrderedDict()  # total trade cost for each trade time
+        self.costs: dict = OrderedDict()  # trade cost rate for each trade time
+        self.values: dict = OrderedDict()  # value for each trade time
+        self.cashes: dict = OrderedDict()
+        self.benches: dict = OrderedDict()
+        self.latest_pm_time: Optional[pd.TimeStamp] = None

-    def init_bench(self, freq=None, benchmark_config=None):
+    def init_bench(self, freq: str = None, benchmark_config: dict = None) -> None:
        if freq is not None:
            self.freq = freq
        self.benchmark_config = benchmark_config
        self.bench = self._cal_benchmark(self.benchmark_config, self.freq)

-    def _cal_benchmark(self, benchmark_config, freq):
+    @staticmethod
+    def _cal_benchmark(benchmark_config: Optional[dict], freq: str) -> Optional[pd.Series]:
        if benchmark_config is None:
            return None
        benchmark = benchmark_config.get("benchmark", CSI300_BENCH)
@@ -110,7 +115,12 @@ class PortfolioMetrics:
                raise ValueError(f"The benchmark {_codes} does not exist. Please provide the right benchmark")
            return _temp_result.groupby(level="datetime")[_temp_result.columns.tolist()[0]].mean().fillna(0)

-    def _sample_benchmark(self, bench, trade_start_time, trade_end_time):
+    def _sample_benchmark(
+        self,
+        bench: pd.Series,
+        trade_start_time: Union[str, pd.Timestamp],
+        trade_end_time: Union[str, pd.Timestamp],
+    ) -> Optional[float]:
        if self.bench is None:
            return None

@@ -120,35 +130,35 @@ class PortfolioMetrics:
        _ret = resam_ts_data(bench, trade_start_time, trade_end_time, method=cal_change)
        return 0.0 if _ret is None else _ret - 1

-    def is_empty(self):
+    def is_empty(self) -> bool:
        return len(self.accounts) == 0

-    def get_latest_date(self):
+    def get_latest_date(self) -> pd.Timestamp:
        return self.latest_pm_time

-    def get_latest_account_value(self):
+    def get_latest_account_value(self) -> float:
        return self.accounts[self.latest_pm_time]

-    def get_latest_total_cost(self):
+    def get_latest_total_cost(self) -> Any:
        return self.total_costs[self.latest_pm_time]

-    def get_latest_total_turnover(self):
+    def get_latest_total_turnover(self) -> Any:
        return self.total_turnovers[self.latest_pm_time]

    def update_portfolio_metrics_record(
        self,
-        trade_start_time=None,
-        trade_end_time=None,
-        account_value=None,
-        cash=None,
-        return_rate=None,
-        total_turnover=None,
-        turnover_rate=None,
-        total_cost=None,
-        cost_rate=None,
-        stock_value=None,
-        bench_value=None,
-    ):
+        trade_start_time: Union[str, pd.Timestamp] = None,
+        trade_end_time: Union[str, pd.Timestamp] = None,
+        account_value: float = None,
+        cash: float = None,
+        return_rate: float = None,
+        total_turnover: float = None,
+        turnover_rate: float = None,
+        total_cost: float = None,
+        cost_rate: float = None,
+        stock_value: float = None,
+        bench_value: float = None,
+    ) -> None:
        # check data
        if None in [
            trade_start_time,
@@ -185,7 +195,7 @@ class PortfolioMetrics:
        self.latest_pm_time = trade_start_time
        # finish pm update in each step

-    def generate_portfolio_metrics_dataframe(self):
+    def generate_portfolio_metrics_dataframe(self) -> pd.DataFrame:
        pm = pd.DataFrame()
        pm["account"] = pd.Series(self.accounts)
        pm["return"] = pd.Series(self.returns)
@@ -199,19 +209,18 @@ class PortfolioMetrics:
        pm.index.name = "datetime"
        return pm

-    def save_portfolio_metrics(self, path):
+    def save_portfolio_metrics(self, path: str) -> None:
        r = self.generate_portfolio_metrics_dataframe()
        r.to_csv(path)

-    def load_portfolio_metrics(self, path):
+    def load_portfolio_metrics(self, path: str) -> None:
        """load pm from a file
        should have format like
        columns = ['account', 'return', 'total_turnover', 'turnover', 'cost', 'total_cost', 'value', 'cash', 'bench']
            :param
                path: str/ pathlib.Path()
        """
-        path = pathlib.Path(path)
-        with path.open("rb") as f:
+        with pathlib.Path(path).open("rb") as f:
            r = pd.read_csv(f, index_col=0)
        r.index = pd.DatetimeIndex(r.index)

@@ -261,30 +270,30 @@ class Indicator:

    """

-    def __init__(self, order_indicator_cls=NumpyOrderIndicator):
+    def __init__(self, order_indicator_cls: Type[BaseOrderIndicator] = NumpyOrderIndicator) -> None:
        self.order_indicator_cls = order_indicator_cls

        # order indicator is metrics for a single order for a specific step
-        self.order_indicator_his = OrderedDict()
+        self.order_indicator_his: dict = OrderedDict()
        self.order_indicator: BaseOrderIndicator = self.order_indicator_cls()

        # trade indicator is metrics for all orders for a specific step
-        self.trade_indicator_his = OrderedDict()
-        self.trade_indicator: Dict[str, float] = OrderedDict()
+        self.trade_indicator_his: dict = OrderedDict()
+        self.trade_indicator: Dict[str, Optional[BaseSingleMetric]] = OrderedDict()

        self._trade_calendar = None

    # def reset(self, trade_calendar: TradeCalendarManager):
-    def reset(self):
-        self.order_indicator: BaseOrderIndicator = self.order_indicator_cls()
+    def reset(self) -> None:
+        self.order_indicator = self.order_indicator_cls()
        self.trade_indicator = OrderedDict()
        # self._trade_calendar = trade_calendar

-    def record(self, trade_start_time):
+    def record(self, trade_start_time: Union[str, pd.Timestamp]) -> None:
        self.order_indicator_his[trade_start_time] = self.get_order_indicator()
        self.trade_indicator_his[trade_start_time] = self.get_trade_indicator()

-    def _update_order_trade_info(self, trade_info: list):
+    def _update_order_trade_info(self, trade_info: List[Tuple[Order, float, float, float]]) -> None:
        amount = dict()
        deal_amount = dict()
        trade_price = dict()
@@ -313,7 +322,7 @@ class Indicator:
        self.order_indicator.assign("trade_dir", trade_dir)
        self.order_indicator.assign("pa", pa)

-    def _update_order_fulfill_rate(self):
+    def _update_order_fulfill_rate(self) -> None:
        def func(deal_amount, amount):
            # deal_amount is np.NaN or None when there is no inner decision. So full fill rate is 0.
            tmp_deal_amount = deal_amount.reindex(amount.index, 0)
@@ -322,11 +331,11 @@ class Indicator:

        self.order_indicator.transfer(func, "ffr")

-    def update_order_indicators(self, trade_info: list):
+    def update_order_indicators(self, trade_info: List[Tuple[Order, float, float, float]]) -> None:
        self._update_order_trade_info(trade_info=trade_info)
        self._update_order_fulfill_rate()

-    def _agg_order_trade_info(self, inner_order_indicators: List[Dict[str, pd.Series]]):
+    def _agg_order_trade_info(self, inner_order_indicators: List[BaseOrderIndicator]) -> None:
        # calculate total trade amount with each inner order indicator.
        def trade_amount_func(deal_amount, trade_price):
            return deal_amount * trade_price
@@ -355,9 +364,9 @@ class Indicator:

        self.order_indicator.transfer(func_apply, "trade_dir")

-    def _update_trade_amount(self, outer_trade_decision: BaseTradeDecision):
+    def _update_trade_amount(self, outer_trade_decision: BaseTradeDecision) -> None:
        # NOTE: these indicator is designed for order execution, so the
-        decision: List[Order] = outer_trade_decision.get_decision()
+        decision: List[Order] = cast(List[Order], outer_trade_decision.get_decision())
        if len(decision) == 0:
            self.order_indicator.assign("amount", {})
        else:
@@ -372,7 +381,7 @@ class Indicator:
        decision: BaseTradeDecision,
        trade_exchange: Exchange,
        pa_config: dict = {},
-    ):
+    ) -> Tuple[Optional[float], Optional[float]]:
        """
        Get the base volume and price information
        All the base price values are rooted from this function
@@ -412,31 +421,35 @@ class Indicator:
        # NOTE: there are some zeros in the trading price. These cases are known meaningless
        # for aligning the previous logic, remove it.
        # remove zero and negative values.
-        price_s = price_s.loc[(price_s > 1e-08).data.astype(np.bool)]
+        assert isinstance(price_s, idd.SingleData)
+        price_s = price_s.loc[(price_s > 1e-08).data.astype(bool)]
        # NOTE ~(price_s < 1e-08) is different from price_s >= 1e-8
        #   ~(np.NaN < 1e-8) -> ~(False)  -> True

+        assert isinstance(price_s, idd.SingleData)
        if agg == "vwap":
            volume_s = trade_exchange.get_volume(inst, trade_start_time, trade_end_time, method=None)
            if isinstance(volume_s, (int, float, np.number)):
                volume_s = idd.SingleData(volume_s, [trade_start_time])
+            assert isinstance(volume_s, idd.SingleData)
            volume_s = volume_s.reindex(price_s.index)
        elif agg == "twap":
            volume_s = idd.SingleData(1, price_s.index)
        else:
            raise NotImplementedError(f"This type of input is not supported")

+        assert isinstance(volume_s, idd.SingleData)
        base_volume = volume_s.sum()
        base_price = (price_s * volume_s).sum() / base_volume
        return base_price, base_volume

    def _agg_base_price(
        self,
-        inner_order_indicators: List[Dict[str, Union[SingleMetric, idd.SingleData]]],
+        inner_order_indicators: List[BaseOrderIndicator],
        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]],
        trade_exchange: Exchange,
        pa_config: dict = {},
-    ):
+    ) -> None:
        """
        # NOTE:!!!!
        # Strong assumption!!!!!!
@@ -444,7 +457,7 @@ class Indicator:

        Parameters
        ----------
-        inner_order_indicators : List[Dict[str, pd.Series]]
+        inner_order_indicators : List[BaseOrderIndicator]
            the indicators of account of inner executor
        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]],
            a list of decisions according to inner_order_indicators
@@ -489,14 +502,17 @@ class Indicator:
                bv_new = idd.SingleData(bv_new)
                bp_all.append(bp_new)
                bv_all.append(bv_new)
-            bp_all = idd.concat(bp_all, axis=1)
-            bv_all = idd.concat(bv_all, axis=1)
+            bp_all_multi_data = idd.concat(bp_all, axis=1)
+            bv_all_multi_data = idd.concat(bv_all, axis=1)

-            base_volume = bv_all.sum(axis=1)
+            base_volume = bv_all_multi_data.sum(axis=1)
            self.order_indicator.assign("base_volume", base_volume.to_dict())
-            self.order_indicator.assign("base_price", ((bp_all * bv_all).sum(axis=1) / base_volume).to_dict())
+            self.order_indicator.assign(
+                "base_price",
+                ((bp_all_multi_data * bv_all_multi_data).sum(axis=1) / base_volume).to_dict(),
+            )

-    def _agg_order_price_advantage(self):
+    def _agg_order_price_advantage(self) -> None:
        def if_empty_func(trade_price):
            return trade_price.empty

@@ -513,12 +529,12 @@ class Indicator:

    def agg_order_indicators(
        self,
-        inner_order_indicators: List[Dict[str, pd.Series]],
+        inner_order_indicators: List[BaseOrderIndicator],
        decision_list: List[Tuple[BaseTradeDecision, pd.Timestamp, pd.Timestamp]],
        outer_trade_decision: BaseTradeDecision,
        trade_exchange: Exchange,
-        indicator_config={},
-    ):
+        indicator_config: dict = {},
+    ) -> None:
        self._agg_order_trade_info(inner_order_indicators)
        self._update_trade_amount(outer_trade_decision)
        self._update_order_fulfill_rate()
@@ -526,71 +542,66 @@ class Indicator:
        self._agg_base_price(inner_order_indicators, decision_list, trade_exchange, pa_config=pa_config)  # TODO
        self._agg_order_price_advantage()

-    def _cal_trade_fulfill_rate(self, method="mean"):
+    def _cal_trade_fulfill_rate(self, method: str = "mean") -> Optional[BaseSingleMetric]:
        if method == "mean":
-
-            def func(ffr):
-                return ffr.mean()
-
+            return self.order_indicator.transfer(
+                lambda ffr: ffr.mean(),
+            )
        elif method == "amount_weighted":
-
-            def func(ffr, deal_amount):
-                return (ffr * deal_amount.abs()).sum() / (deal_amount.abs().sum())
-
+            return self.order_indicator.transfer(
+                lambda ffr, deal_amount: (ffr * deal_amount.abs()).sum() / (deal_amount.abs().sum()),
+            )
        elif method == "value_weighted":
-
-            def func(ffr, trade_value):
-                return (ffr * trade_value.abs()).sum() / (trade_value.abs().sum())
-
+            return self.order_indicator.transfer(
+                lambda ffr, trade_value: (ffr * trade_value.abs()).sum() / (trade_value.abs().sum()),
+            )
        else:
            raise ValueError(f"method {method} is not supported!")
-        return self.order_indicator.transfer(func)

-    def _cal_trade_price_advantage(self, method="mean"):
+    def _cal_trade_price_advantage(self, method: str = "mean") -> Optional[BaseSingleMetric]:
        if method == "mean":
-
-            def func(pa):
-                return pa.mean()
-
+            return self.order_indicator.transfer(lambda pa: pa.mean())
        elif method == "amount_weighted":
-
-            def func(pa, deal_amount):
-                return (pa * deal_amount.abs()).sum() / (deal_amount.abs().sum())
-
+            return self.order_indicator.transfer(
+                lambda pa, deal_amount: (pa * deal_amount.abs()).sum() / (deal_amount.abs().sum()),
+            )
        elif method == "value_weighted":
-
-            def func(pa, trade_value):
-                return (pa * trade_value.abs()).sum() / (trade_value.abs().sum())
-
+            return self.order_indicator.transfer(
+                lambda pa, trade_value: (pa * trade_value.abs()).sum() / (trade_value.abs().sum()),
+            )
        else:
            raise ValueError(f"method {method} is not supported!")
-        return self.order_indicator.transfer(func)

-    def _cal_trade_positive_rate(self):
+    def _cal_trade_positive_rate(self) -> Optional[BaseSingleMetric]:
        def func(pa):
            return (pa > 0).sum() / pa.count()

        return self.order_indicator.transfer(func)

-    def _cal_deal_amount(self):
+    def _cal_deal_amount(self) -> Optional[BaseSingleMetric]:
        def func(deal_amount):
            return deal_amount.abs().sum()

        return self.order_indicator.transfer(func)

-    def _cal_trade_value(self):
+    def _cal_trade_value(self) -> Optional[BaseSingleMetric]:
        def func(trade_value):
            return trade_value.abs().sum()

        return self.order_indicator.transfer(func)

-    def _cal_trade_order_count(self):
+    def _cal_trade_order_count(self) -> Optional[BaseSingleMetric]:
        def func(amount):
            return amount.count()

        return self.order_indicator.transfer(func)

-    def cal_trade_indicators(self, trade_start_time, freq, indicator_config={}):
+    def cal_trade_indicators(
+        self,
+        trade_start_time: Union[str, pd.Timestamp],
+        freq: str,
+        indicator_config: dict = {},
+    ) -> None:
        show_indicator = indicator_config.get("show_indicator", False)
        ffr_config = indicator_config.get("ffr_config", {})
        pa_config = indicator_config.get("pa_config", {})
@@ -608,22 +619,22 @@ class Indicator:
        self.trade_indicator["count"] = order_count
        if show_indicator:
            print(
-                "[Indicator({}) {:%Y-%m-%d %H:%M:%S}]: FFR: {}, PA: {}, POS: {}".format(
+                "[Indicator({}) {}]: FFR: {}, PA: {}, POS: {}".format(
                    freq,
-                    trade_start_time,
+                    trade_start_time
+                    if isinstance(trade_start_time, str)
+                    else trade_start_time.strftime("%Y-%m-%d %H:%M:%S"),
                    fulfill_rate,
                    price_advantage,
                    positive_rate,
                ),
            )

-    def get_order_indicator(self, raw: bool = True):
-        if raw:
-            return self.order_indicator
-        return self.order_indicator.to_series()
+    def get_order_indicator(self, raw: bool = True) -> Union[BaseOrderIndicator, Dict[Text, pd.Series]]:
+        return self.order_indicator if raw else self.order_indicator.to_series()

-    def get_trade_indicator(self):
+    def get_trade_indicator(self) -> Dict[str, Optional[BaseSingleMetric]]:
        return self.trade_indicator

-    def generate_trade_indicators_dataframe(self):
+    def generate_trade_indicators_dataframe(self) -> pd.DataFrame:
        return pd.DataFrame.from_dict(self.trade_indicator_his, orient="index")
--- a/qlib/backtest/signal.py
+++ b/qlib/backtest/signal.py
@@ -22,7 +22,7 @@ class Signal(metaclass=abc.ABCMeta):
    """

    @abc.abstractmethod
-    def get_signal(self, start_time, end_time) -> Union[pd.Series, pd.DataFrame, None]:
+    def get_signal(self, start_time: pd.Timestamp, end_time: pd.Timestamp) -> Union[pd.Series, pd.DataFrame, None]:
        """
        get the signal at the end of the decision step(from `start_time` to `end_time`)

@@ -39,13 +39,14 @@ class SignalWCache(Signal):
    SignalWCache will store the prepared signal as a attribute and give the according signal based on input query
    """

-    def __init__(self, signal: Union[pd.Series, pd.DataFrame]):
+    def __init__(self, signal: Union[pd.Series, pd.DataFrame]) -> None:
        """

        Parameters
        ----------
        signal : Union[pd.Series, pd.DataFrame]
-            The expected format of the signal is like the data below (the order of index is not important and can be automatically adjusted)
+            The expected format of the signal is like the data below (the order of index is not important and can be
+            automatically adjusted)

                instrument datetime
                SH600000   2008-01-02  0.079704
@@ -56,8 +57,8 @@ class SignalWCache(Signal):
        """
        self.signal_cache = convert_index_format(signal, level="datetime")

-    def get_signal(self, start_time, end_time) -> Union[pd.Series, pd.DataFrame]:
-        # the frequency of the signal may not algin with the decision frequency of strategy
+    def get_signal(self, start_time: pd.Timestamp, end_time: pd.Timestamp) -> Union[pd.Series, pd.DataFrame]:
+        # the frequency of the signal may not align with the decision frequency of strategy
        # so resampling from the data is necessary
        # the latest signal leverage more recent data and therefore is used in trading.
        signal = resam_ts_data(self.signal_cache, start_time=start_time, end_time=end_time, method="last")
@@ -65,7 +66,7 @@ class SignalWCache(Signal):


 class ModelSignal(SignalWCache):
-    def __init__(self, model: BaseModel, dataset: Dataset):
+    def __init__(self, model: BaseModel, dataset: Dataset) -> None:
        self.model = model
        self.dataset = dataset
        pred_scores = self.model.predict(dataset)
@@ -73,7 +74,7 @@ class ModelSignal(SignalWCache):
            pred_scores = pred_scores.iloc[:, 0]
        super().__init__(pred_scores)

-    def _update_model(self):
+    def _update_model(self) -> None:
        """
        When using online data, update model in each bar as the following steps:
            - update dataset with online data, the dataset should support online update
--- a/qlib/backtest/utils.py
+++ b/qlib/backtest/utils.py
@@ -149,6 +149,8 @@ class TradeCalendarManager:
        Tuple[int, int]:
        """
        # potential performance issue
+        assert self.level_infra is not None
+
        day_start = pd.Timestamp(self.start_time.date())
        day_end = epsilon_change(day_start + pd.Timedelta(days=1))
        freq = self.level_infra.get("common_infra").get("trade_exchange").freq
@@ -182,8 +184,8 @@ class TradeCalendarManager:
        Tuple[int, int]:
            the index of the range.  **the left and right are closed**
        """
-        left = bisect.bisect_right(self._calendar, start_time) - 1
-        right = bisect.bisect_right(self._calendar, end_time) - 1
+        left = bisect.bisect_right(list(self._calendar), start_time) - 1
+        right = bisect.bisect_right(list(self._calendar), end_time) - 1
        left -= self.start_index
        right -= self.start_index

@@ -201,14 +203,14 @@ class TradeCalendarManager:


 class BaseInfrastructure:
-    def __init__(self, **kwargs) -> None:
+    def __init__(self, **kwargs: Any) -> None:
        self.reset_infra(**kwargs)

    @abstractmethod
    def get_support_infra(self) -> Set[str]:
        raise NotImplementedError("`get_support_infra` is not implemented!")

-    def reset_infra(self, **kwargs) -> None:
+    def reset_infra(self, **kwargs: Any) -> None:
        support_infra = self.get_support_infra()
        for k, v in kwargs.items():
            if k in support_infra:
--- a/qlib/contrib/evaluate.py
+++ b/qlib/contrib/evaluate.py
@@ -339,7 +339,7 @@ def long_short_backtest(
        for stock in long_stocks:
            if not trade_exchange.is_stock_tradable(stock_id=stock, trade_date=date):
                continue
-            profit = trade_exchange.get_quote_info(stock_id=stock, trade_date=date)[profit_str]
+            profit = trade_exchange.get_quote_info(stock_id=stock, start_time=date, end_time=date, field=profit_str)
            if np.isnan(profit):
                long_profit.append(0)
            else:
@@ -348,17 +348,17 @@ def long_short_backtest(
        for stock in short_stocks:
            if not trade_exchange.is_stock_tradable(stock_id=stock, trade_date=date):
                continue
-            profit = trade_exchange.get_quote_info(stock_id=stock, trade_date=date)[profit_str]
+            profit = trade_exchange.get_quote_info(stock_id=stock, start_time=date, end_time=date, field=profit_str)
            if np.isnan(profit):
                short_profit.append(0)
            else:
-                short_profit.append(-profit)
+                short_profit.append(profit * -1)

        for stock in list(score.loc(axis=0)[pdate, :].index.get_level_values(level=0)):
            # exclude the suspend stock
            if trade_exchange.check_stock_suspended(stock_id=stock, trade_date=date):
                continue
-            profit = trade_exchange.get_quote_info(stock_id=stock, trade_date=date)[profit_str]
+            profit = trade_exchange.get_quote_info(stock_id=stock, start_time=date, end_time=date, field=profit_str)
            if np.isnan(profit):
                all_profit.append(0)
            else:
--- a/qlib/data/data.py
+++ b/qlib/data/data.py
@@ -108,14 +108,16 @@ class CalendarProvider(abc.ABC):
        _, _, si, ei = self.locate_index(start_time, end_time, freq, future)
        return _calendar[si : ei + 1]

-    def locate_index(self, start_time, end_time, freq, future=False):
+    def locate_index(
+        self, start_time: Union[pd.Timestamp, str], end_time: Union[pd.Timestamp, str], freq: str, future: bool = False
+    ):
        """Locate the start time index and end time index in a calendar under certain frequency.

        Parameters
        ----------
-        start_time : str
+        start_time : pd.Timestamp
            start of the time range.
-        end_time : str
+        end_time : pd.Timestamp
            end of the time range.
        freq : str
            time frequency, available: year/quarter/month/week/day.
--- a/qlib/data/storage/file_storage.py
+++ b/qlib/data/storage/file_storage.py
@@ -106,10 +106,7 @@ class FileCalendarStorage(FileStorageMixin, CalendarStorage):
        if not self.uri.exists():
            self._write_calendar(values=[])
        with self.uri.open("rb") as fp:
-            return [
-                str(x)
-                for x in np.loadtxt(fp, str, skiprows=skip_rows, max_rows=n_rows, delimiter="\n", encoding="utf-8")
-            ]
+            return [str(x) for x in np.loadtxt(fp, str, skiprows=skip_rows, max_rows=n_rows, encoding="utf-8")]

    def _write_calendar(self, values: Iterable[CalVT], mode: str = "wb"):
        with self.uri.open(mode=mode) as fp:
--- a/qlib/model/trainer.py
+++ b/qlib/model/trainer.py
@@ -12,7 +12,7 @@ In ``DelayTrainer``, the first step is only to save some necessary info to model
 """

 import socket
-from typing import Callable, List
+from typing import Callable, List, Optional

 from tqdm.auto import tqdm

@@ -219,7 +219,13 @@ class TrainerR(Trainer):
    STATUS_BEGIN = "begin_task_train"
    STATUS_END = "end_task_train"

-    def __init__(self, experiment_name: str = None, train_func: Callable = task_train, call_in_subproc: bool = False):
+    def __init__(
+        self,
+        experiment_name: Optional[str] = None,
+        train_func: Callable = task_train,
+        call_in_subproc: bool = False,
+        default_rec_name: Optional[str] = None,
+    ):
        """
        Init TrainerR.

@@ -230,6 +236,7 @@ class TrainerR(Trainer):
        """
        super().__init__()
        self.experiment_name = experiment_name
+        self.default_rec_name = default_rec_name
        self.train_func = train_func
        self._call_in_subproc = call_in_subproc

@@ -259,7 +266,7 @@ class TrainerR(Trainer):
            if self._call_in_subproc:
                get_module_logger("TrainerR").info("running models in sub process (for forcing release memroy).")
                train_func = call_in_subproc(train_func, C)
-            rec = train_func(task, experiment_name, **kwargs)
+            rec = train_func(task, experiment_name, recorder_name=self.default_rec_name, **kwargs)
            rec.set_tags(**{self.STATUS_KEY: self.STATUS_BEGIN})
            recs.append(rec)
        return recs
@@ -286,7 +293,9 @@ class DelayTrainerR(TrainerR):
    A delayed implementation based on TrainerR, which means `train` method may only do some preparation and `end_train` method can do the real model fitting.
    """

-    def __init__(self, experiment_name: str = None, train_func=begin_task_train, end_train_func=end_task_train):
+    def __init__(
+        self, experiment_name: str = None, train_func=begin_task_train, end_train_func=end_task_train, **kwargs
+    ):
        """
        Init TrainerRM.

@@ -295,7 +304,7 @@ class DelayTrainerR(TrainerR):
            train_func (Callable, optional): default train method. Defaults to `begin_task_train`.
            end_train_func (Callable, optional): default end_train method. Defaults to `end_task_train`.
        """
-        super().__init__(experiment_name, train_func)
+        super().__init__(experiment_name, train_func, **kwargs)
        self.end_train_func = end_train_func
        self.delay = True

@@ -344,7 +353,12 @@ class TrainerRM(Trainer):
    TM_ID = "_id in TaskManager"

    def __init__(
-        self, experiment_name: str = None, task_pool: str = None, train_func=task_train, skip_run_task: bool = False
+        self,
+        experiment_name: str = None,
+        task_pool: str = None,
+        train_func=task_train,
+        skip_run_task: bool = False,
+        default_rec_name: Optional[str] = None,
    ):
        """
        Init TrainerR.
@@ -363,6 +377,7 @@ class TrainerRM(Trainer):
        self.task_pool = task_pool
        self.train_func = train_func
        self.skip_run_task = skip_run_task
+        self.default_rec_name = default_rec_name

    def train(
        self,
@@ -371,6 +386,7 @@ class TrainerRM(Trainer):
        experiment_name: str = None,
        before_status: str = TaskManager.STATUS_WAITING,
        after_status: str = TaskManager.STATUS_DONE,
+        default_rec_name: Optional[str] = None,
        **kwargs,
    ) -> List[Recorder]:
        """
@@ -398,6 +414,8 @@ class TrainerRM(Trainer):
            train_func = self.train_func
        if experiment_name is None:
            experiment_name = self.experiment_name
+        if default_rec_name is None:
+            default_rec_name = self.default_rec_name
        task_pool = self.task_pool
        if task_pool is None:
            task_pool = experiment_name
@@ -412,6 +430,7 @@ class TrainerRM(Trainer):
                experiment_name=experiment_name,
                before_status=before_status,
                after_status=after_status,
+                recorder_name=default_rec_name,
                **kwargs,
            )

@@ -480,6 +499,7 @@ class DelayTrainerRM(TrainerRM):
        train_func=begin_task_train,
        end_train_func=end_task_train,
        skip_run_task: bool = False,
+        **kwargs,
    ):
        """
        Init DelayTrainerRM.
@@ -494,7 +514,7 @@ class DelayTrainerRM(TrainerRM):
                Only run_task in the worker. Otherwise skip run_task.
                E.g. Starting trainer on a CPU VM and then waiting tasks to be finished on GPU VMs.
        """
-        super().__init__(experiment_name, task_pool, train_func)
+        super().__init__(experiment_name, task_pool, train_func, **kwargs)
        self.end_train_func = end_train_func
        self.delay = True
        self.skip_run_task = skip_run_task
--- a/qlib/rl/data/pickle_styled.py
+++ b/qlib/rl/data/pickle_styled.py
@@ -248,7 +248,7 @@ def load_orders(
            Order(
                row["instrument"],
                row["amount"],
-                int(row["order_type"]),
+                OrderDir(int(row["order_type"])),
                row["datetime"].replace(hour=start_time.hour, minute=start_time.minute, second=start_time.second),
                row["datetime"].replace(hour=end_time.hour, minute=end_time.minute, second=end_time.second),
            )
--- a/qlib/rl/entries/init.py
+++ b/qlib/rl/entries/init.py
@@ -1,7 +0,0 @@
-# Copyright (c) Microsoft Corporation.
-# Licensed under the MIT License.
-
-"""Train, test, inference utilities.
-
-The APIs in this directory are NOT considered final and are subject to change!
-"""
--- a/qlib/rl/entries/test.py
+++ b/qlib/rl/entries/test.py
@@ -1,99 +0,0 @@
-# Copyright (c) Microsoft Corporation.
-# Licensed under the MIT License.
-
-from __future__ import annotations
-
-import copy
-from typing import Callable, Sequence
-
-from tianshou.data import Collector
-from tianshou.policy import BasePolicy
-
-from qlib.constant import INF
-from qlib.log import get_module_logger
-from qlib.rl.simulator import InitialStateType, Simulator
-from qlib.rl.interpreter import StateInterpreter, ActionInterpreter
-from qlib.rl.reward import Reward
-from qlib.rl.utils import DataQueue, EnvWrapper, FiniteEnvType, LogCollector, LogWriter, vectorize_env
-
-
-_logger = get_module_logger(__name__)
-
-
-def backtest(
-    simulator_fn: Callable[[InitialStateType], Simulator],
-    state_interpreter: StateInterpreter,
-    action_interpreter: ActionInterpreter,
-    initial_states: Sequence[InitialStateType],
-    policy: BasePolicy,
-    logger: LogWriter | list[LogWriter],
-    reward: Reward | None = None,
-    finite_env_type: FiniteEnvType = "subproc",
-    concurrency: int = 2,
-) -> None:
-    """Backtest with the parallelism provided by RL framework.
-
-    Parameters
-    ----------
-    simulator_fn
-        Callable receiving initial seed, returning a simulator.
-    state_interpreter
-        Interprets the state of simulators.
-    action_interpreter
-        Interprets the policy actions.
-    initial_states
-        Initial states to iterate over. Every state will be run exactly once.
-    policy
-        Policy to test against.
-    logger
-        Logger to record the backtest results. Logger must be present because
-        without logger, all information will be lost.
-    reward
-        Optional reward function. For backtest, this is for testing the rewards
-        and logging them only.
-    finite_env_type
-        Type of finite env implementation.
-    concurrency
-        Parallel workers.
-    """
-
-    # To save bandwidth
-    min_loglevel = min(lg.loglevel for lg in logger) if isinstance(logger, list) else logger.loglevel
-
-    def env_factory():
-        # FIXME: state_interpreter and action_interpreter are stateful (having a weakref of env),
-        # and could be thread unsafe.
-        # I'm not sure whether it's a design flaw.
-        # I'll rethink about this when designing the trainer.
-
-        if finite_env_type == "dummy":
-            # We could only experience the "threading-unsafe" problem in dummy.
-            state = copy.deepcopy(state_interpreter)
-            action = copy.deepcopy(action_interpreter)
-            rew = copy.deepcopy(reward)
-        else:
-            state, action, rew = state_interpreter, action_interpreter, reward
-
-        return EnvWrapper(
-            simulator_fn,
-            state,
-            action,
-            seed_iterator,
-            rew,
-            logger=LogCollector(min_loglevel=min_loglevel),
-        )
-
-    with DataQueue(initial_states) as seed_iterator:
-        vector_env = vectorize_env(
-            env_factory,
-            finite_env_type,
-            concurrency,
-            logger,
-        )
-
-        policy.eval()
-
-        with vector_env.collector_guard():
-            test_collector = Collector(policy, vector_env)
-            _logger.info("All ready. Start backtest.")
-            test_collector.collect(n_step=INF * len(vector_env))
--- a/qlib/rl/entries/train.py
+++ b/qlib/rl/entries/train.py
@@ -1,4 +0,0 @@
-# Copyright (c) Microsoft Corporation.
-# Licensed under the MIT License.
-
-# TBD
--- a/qlib/rl/order_execution/init.py
+++ b/qlib/rl/order_execution/init.py
@@ -9,4 +9,5 @@ Multi-asset is on the way.
 from .interpreter import *
 from .network import *
 from .policy import *
+from .reward import *
 from .simulator_simple import *
--- a/qlib/rl/order_execution/reward.py
+++ b/qlib/rl/order_execution/reward.py
@@ -0,0 +1,46 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from __future__ import annotations
+
+from typing import cast
+
+import numpy as np
+from qlib.rl.reward import Reward
+
+from .simulator_simple import SAOEState, SAOEMetrics
+
+__all__ = ["PAPenaltyReward"]
+
+
+class PAPenaltyReward(Reward[SAOEState]):
+    """Encourage higher PAs, but penalize stacking all the amounts within a very short time.
+    Formally, for each time step, the reward is :math:`(PA_t * vol_t / target - vol_t^2 * penalty)`.
+
+    Parameters
+    ----------
+    penalty
+        The penalty for large volume in a short time.
+    """
+
+    def __init__(self, penalty: float = 100.0):
+        self.penalty = penalty
+
+    def reward(self, simulator_state: SAOEState) -> float:
+        whole_order = simulator_state.order.amount
+        assert whole_order > 0
+        last_step = cast(SAOEMetrics, simulator_state.history_steps.reset_index().iloc[-1].to_dict())
+        pa = last_step["pa"] * last_step["amount"] / whole_order
+
+        # Inspect the "break-down" of the latest step: trading amount at every tick
+        last_step_breakdown = simulator_state.history_exec.loc[last_step["datetime"] :]
+        penalty = -self.penalty * ((last_step_breakdown["amount"] / whole_order) ** 2).sum()
+
+        reward = pa + penalty
+
+        # Throw error in case of NaN
+        assert not (np.isnan(reward) or np.isinf(reward)), f"Invalid reward for simulator state: {simulator_state}"
+
+        self.log("reward/pa", pa)
+        self.log("reward/penalty", penalty)
+        return reward
--- a/qlib/rl/order_execution/simulator_simple.py
+++ b/qlib/rl/order_execution/simulator_simple.py
@@ -131,11 +131,14 @@ class SingleAssetOrderExecution(Simulator[Order, SAOEState, float]):
    """

    history_exec: pd.DataFrame
-    """All execution history at every possible time ticks. See :class:`SAOEMetrics` for available columns."""
+    """All execution history at every possible time ticks. See :class:`SAOEMetrics` for available columns.
+    Index is ``datetime``.
+    """

    history_steps: pd.DataFrame
    """Positions at each step. The position before first step is also recorded.
-    See :class:`SAOEMetrics` for available columns."""
+    See :class:`SAOEMetrics` for available columns.
+    Index is ``datetime``, which is the **starting** time of each step."""

    metrics: SAOEMetrics | None
    """Metrics. Only available when done."""
--- a/qlib/rl/trainer/init.py
+++ b/qlib/rl/trainer/init.py
@@ -0,0 +1,9 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+"""Train, test, inference utilities."""
+
+from .api import backtest, train
+from .callbacks import EarlyStopping, Checkpoint
+from .trainer import Trainer
+from .vessel import TrainingVessel, TrainingVesselBase
--- a/qlib/rl/trainer/api.py
+++ b/qlib/rl/trainer/api.py
@@ -0,0 +1,118 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from __future__ import annotations
+
+from typing import Callable, Sequence, cast, Any
+
+from tianshou.policy import BasePolicy
+
+from qlib.rl.simulator import InitialStateType, Simulator
+from qlib.rl.interpreter import StateInterpreter, ActionInterpreter
+from qlib.rl.reward import Reward
+from qlib.rl.utils import FiniteEnvType, LogWriter
+
+from .vessel import TrainingVessel
+from .trainer import Trainer
+
+
+def train(
+    simulator_fn: Callable[[InitialStateType], Simulator],
+    state_interpreter: StateInterpreter,
+    action_interpreter: ActionInterpreter,
+    initial_states: Sequence[InitialStateType],
+    policy: BasePolicy,
+    reward: Reward,
+    vessel_kwargs: dict[str, Any],
+    trainer_kwargs: dict[str, Any],
+) -> None:
+    """Train a policy with the parallelism provided by RL framework.
+
+    Experimental API. Parameters might change shortly.
+
+    Parameters
+    ----------
+    simulator_fn
+        Callable receiving initial seed, returning a simulator.
+    state_interpreter
+        Interprets the state of simulators.
+    action_interpreter
+        Interprets the policy actions.
+    initial_states
+        Initial states to iterate over. Every state will be run exactly once.
+    policy
+        Policy to train against.
+    reward
+        Reward function.
+    vessel_kwargs
+        Keyword arguments passed to :class:`TrainingVessel`, like ``episode_per_iter``.
+    trainer_kwargs
+        Keyword arguments passed to :class:`Trainer`, like ``finite_env_type``, ``concurrency``.
+    """
+
+    vessel = TrainingVessel(
+        simulator_fn=simulator_fn,
+        state_interpreter=state_interpreter,
+        action_interpreter=action_interpreter,
+        policy=policy,
+        train_initial_states=initial_states,
+        reward=reward,  # ignore none
+        **vessel_kwargs,
+    )
+    trainer = Trainer(**trainer_kwargs)
+    trainer.fit(vessel)
+
+
+def backtest(
+    simulator_fn: Callable[[InitialStateType], Simulator],
+    state_interpreter: StateInterpreter,
+    action_interpreter: ActionInterpreter,
+    initial_states: Sequence[InitialStateType],
+    policy: BasePolicy,
+    logger: LogWriter | list[LogWriter],
+    reward: Reward | None = None,
+    finite_env_type: FiniteEnvType = "subproc",
+    concurrency: int = 2,
+) -> None:
+    """Backtest with the parallelism provided by RL framework.
+
+    Experimental API. Parameters might change shortly.
+
+    Parameters
+    ----------
+    simulator_fn
+        Callable receiving initial seed, returning a simulator.
+    state_interpreter
+        Interprets the state of simulators.
+    action_interpreter
+        Interprets the policy actions.
+    initial_states
+        Initial states to iterate over. Every state will be run exactly once.
+    policy
+        Policy to test against.
+    logger
+        Logger to record the backtest results. Logger must be present because
+        without logger, all information will be lost.
+    reward
+        Optional reward function. For backtest, this is for testing the rewards
+        and logging them only.
+    finite_env_type
+        Type of finite env implementation.
+    concurrency
+        Parallel workers.
+    """
+
+    vessel = TrainingVessel(
+        simulator_fn=simulator_fn,
+        state_interpreter=state_interpreter,
+        action_interpreter=action_interpreter,
+        policy=policy,
+        test_initial_states=initial_states,
+        reward=cast(Reward, reward),  # ignore none
+    )
+    trainer = Trainer(
+        finite_env_type=finite_env_type,
+        concurrency=concurrency,
+        loggers=logger,
+    )
+    trainer.test(vessel)
--- a/qlib/rl/trainer/callbacks.py
+++ b/qlib/rl/trainer/callbacks.py
@@ -0,0 +1,267 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+"""Callbacks to insert customized recipes during the training.
+Mimicks the hooks of Keras / PyTorch-Lightning, but tailored for the context of RL.
+"""
+
+from __future__ import annotations
+
+import copy
+import shutil
+import time
+from datetime import datetime
+from pathlib import Path
+from typing import Any, TYPE_CHECKING
+
+import numpy as np
+import torch
+
+from qlib.log import get_module_logger
+from qlib.typehint import Literal
+
+if TYPE_CHECKING:
+    from .trainer import Trainer
+    from .vessel import TrainingVesselBase
+
+
+_logger = get_module_logger(__name__)
+
+
+class Callback:
+    """Base class of all callbacks."""
+
+    def on_fit_start(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called before the whole fit process begins."""
+
+    def on_fit_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called after the whole fit process ends."""
+
+    def on_train_start(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when each collect for training begins."""
+
+    def on_train_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when the training ends.
+        To access all outputs produced during training, cache the data in either trainer and vessel,
+        and post-process them in this hook.
+        """
+
+    def on_validate_start(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when every run for validation begins."""
+
+    def on_validate_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when the validation ends."""
+
+    def on_test_start(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when every run of testing begins."""
+
+    def on_test_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when the testing ends."""
+
+    def on_iter_start(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called when every iteration (i.e., collect) starts."""
+
+    def on_iter_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        """Called upon every end of iteration.
+        This is called **after** the bump of ``current_iter``,
+        when the previous iteration is considered complete.
+        """
+
+    def state_dict(self) -> Any:
+        """Get a state dict of the callback for pause and resume."""
+
+    def load_state_dict(self, state_dict: Any) -> None:
+        """Resume the callback from a saved state dict."""
+
+
+class EarlyStopping(Callback):
+    """Stop training when a monitored metric has stopped improving.
+
+    The earlystopping callback will be triggered each time validation ends.
+    It will examine the metrics produced in validation,
+    and get the metric with name ``monitor` (``monitor`` is ``reward`` by default),
+    to check whether it's no longer increasing / decreasing.
+    It takes ``min_delta`` and ``patience`` if applicable.
+    If it's found to be not increasing / decreasing any more.
+    ``trainer.should_stop`` will be set to true,
+    and the training terminates.
+
+    Implementation reference: https://github.com/keras-team/keras/blob/v2.9.0/keras/callbacks.py#L1744-L1893
+    """
+
+    def __init__(
+        self,
+        monitor: str = "reward",
+        min_delta: float = 0.0,
+        patience: int = 0,
+        mode: Literal["min", "max"] = "max",
+        baseline: float | None = None,
+        restore_best_weights: bool = False,
+    ):
+        super().__init__()
+
+        self.monitor = monitor
+        self.patience = patience
+        self.baseline = baseline
+        self.min_delta = abs(min_delta)
+        self.restore_best_weights = restore_best_weights
+        self.best_weights: Any | None = None
+
+        if mode not in ["min", "max"]:
+            raise ValueError("Unsupported earlystopping mode: " + mode)
+
+        if mode == "min":
+            self.monitor_op = np.less
+        elif mode == "max":
+            self.monitor_op = np.greater
+
+        if self.monitor_op == np.greater:
+            self.min_delta *= 1
+        else:
+            self.min_delta *= -1
+
+    def state_dict(self) -> dict:
+        return {"wait": self.wait, "best": self.best, "best_weights": self.best_weights, "best_iter": self.best_iter}
+
+    def load_state_dict(self, state_dict: dict) -> None:
+        self.wait = state_dict["wait"]
+        self.best = state_dict["best"]
+        self.best_weights = state_dict["best_weights"]
+        self.best_iter = state_dict["best_iter"]
+
+    def on_fit_start(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        # Allow instances to be re-used
+        self.wait = 0
+        self.best = np.inf if self.monitor_op == np.less else -np.inf
+        self.best_weights = None
+        self.best_iter = 0
+
+    def on_validate_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        current = self.get_monitor_value(trainer)
+        if current is None:
+            return
+        if self.restore_best_weights and self.best_weights is None:
+            # Restore the weights after first iteration if no progress is ever made.
+            self.best_weights = copy.deepcopy(vessel.state_dict())
+
+        self.wait += 1
+        if self._is_improvement(current, self.best):
+            self.best = current
+            self.best_iter = trainer.current_iter
+            if self.restore_best_weights:
+                self.best_weights = copy.deepcopy(vessel.state_dict())
+            # Only restart wait if we beat both the baseline and our previous best.
+            if self.baseline is None or self._is_improvement(current, self.baseline):
+                self.wait = 0
+
+        # Only check after the first epoch.
+        if self.wait >= self.patience and trainer.current_iter > 0:
+            trainer.should_stop = True
+            _logger.info(f"On iteration %d: early stopping", trainer.current_iter + 1)
+            if self.restore_best_weights and self.best_weights is not None:
+                _logger.info("Restoring model weights from the end of the best iteration: %d", self.best_iter + 1)
+                vessel.load_state_dict(self.best_weights)
+
+    def get_monitor_value(self, trainer: Trainer) -> Any:
+        monitor_value = trainer.metrics.get(self.monitor)
+        if monitor_value is None:
+            _logger.warning(
+                "Early stopping conditioned on metric `%s` which is not available. Available metrics are: %s",
+                self.monitor,
+                ",".join(list(trainer.metrics.keys())),
+            )
+        return monitor_value
+
+    def _is_improvement(self, monitor_value, reference_value):
+        return self.monitor_op(monitor_value - self.min_delta, reference_value)
+
+
+class Checkpoint(Callback):
+    """Save checkpoints periodically for persistence and recovery.
+
+    Reference: https://github.com/PyTorchLightning/pytorch-lightning/blob/bfa8b7be/pytorch_lightning/callbacks/model_checkpoint.py
+
+    Parameters
+    ----------
+    dirpath
+        Directory to save the checkpoint file.
+    filename
+        Checkpoint filename. Can contain named formatting options to be auto-filled.
+        For example: ``{iter:03d}-{reward:.2f}.pth``.
+        Supported argument names are:
+
+        - iter (int)
+        - metrics in ``trainer.metrics``
+        - time string, in the format of ``%Y%m%d%H%M%S``
+    save_latest
+        Save the latest checkpoint in ``latest.pth``.
+        If ``link``, ``latest.pth`` will be created as a softlink.
+        If ``copy``, ``latest.pth`` will be stored as an individual copy.
+        Set to none to disable this.
+    every_n_iters
+        Checkpoints are saved at the end of every n iterations of training,
+        after validation if applicable.
+    time_interval
+        Maximum time (seconds) before checkpoints save again.
+    save_on_fit_end
+        Save one last checkpoint at the end to fit.
+        Do nothing if a checkpoint is already saved there.
+    """
+
+    def __init__(
+        self,
+        dirpath: Path,
+        filename: str = "{iter:03d}.pth",
+        save_latest: Literal["link", "copy"] | None = "link",
+        every_n_iters: int | None = None,
+        time_interval: int | None = None,
+        save_on_fit_end: bool = True,
+    ):
+        self.dirpath = Path(dirpath)
+        self.filename = filename
+        self.save_latest = save_latest
+        self.every_n_iters = every_n_iters
+        self.time_interval = time_interval
+        self.save_on_fit_end = save_on_fit_end
+
+        self._last_checkpoint_name: str | None = None
+        self._last_checkpoint_iter: int | None = None
+        self._last_checkpoint_time: float | None = None
+
+    def on_fit_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        if self.save_on_fit_end and (trainer.current_iter != self._last_checkpoint_iter):
+            self._save_checkpoint(trainer)
+
+    def on_iter_end(self, trainer: Trainer, vessel: TrainingVesselBase) -> None:
+        should_save_ckpt = False
+        if self.every_n_iters is not None and (trainer.current_iter + 1) % self.every_n_iters == 0:
+            should_save_ckpt = True
+        if self.time_interval is not None and (
+            self._last_checkpoint_time is None or (time.time() - self._last_checkpoint_time) >= self.time_interval
+        ):
+            should_save_ckpt = True
+        if should_save_ckpt:
+            self._save_checkpoint(trainer)
+
+    def _save_checkpoint(self, trainer: Trainer) -> None:
+        self.dirpath.mkdir(exist_ok=True, parents=True)
+        self._last_checkpoint_name = self._new_checkpoint_name(trainer)
+        self._last_checkpoint_iter = trainer.current_iter
+        self._last_checkpoint_time = time.time()
+        torch.save(trainer.state_dict(), self.dirpath / self._last_checkpoint_name)
+
+        latest_pth = self.dirpath / "latest.pth"
+
+        # Remove first before saving
+        if self.save_latest and latest_pth.exists():
+            latest_pth.unlink()
+
+        if self.save_latest == "link":
+            latest_pth.symlink_to(self.dirpath / self._last_checkpoint_name)
+        elif self.save_latest == "copy":
+            shutil.copyfile(self.dirpath / self._last_checkpoint_name, latest_pth)
+
+    def _new_checkpoint_name(self, trainer: Trainer) -> str:
+        return self.filename.format(
+            iter=trainer.current_iter, time=datetime.now().strftime("%Y%m%d%H%M%S"), **trainer.metrics
+        )
--- a/qlib/rl/trainer/trainer.py
+++ b/qlib/rl/trainer/trainer.py
@@ -0,0 +1,343 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from __future__ import annotations
+
+import copy
+from contextlib import AbstractContextManager, contextmanager
+from pathlib import Path
+from typing import Any, Iterable, TypeVar, Sequence, cast
+
+import torch
+
+from qlib.rl.simulator import InitialStateType
+from qlib.rl.utils import EnvWrapper, FiniteEnvType, LogCollector, LogWriter, LogBuffer, vectorize_env, LogLevel
+from qlib.log import get_module_logger
+from qlib.rl.utils.finite_env import FiniteVectorEnv
+from qlib.typehint import Literal
+
+from .callbacks import Callback
+from .vessel import TrainingVesselBase
+
+_logger = get_module_logger(__name__)
+
+
+T = TypeVar("T")
+
+
+class Trainer:
+    """
+    Utility to train a policy on a particular task.
+
+    Different from traditional DL trainer, the iteration of this trainer is "collect",
+    rather than "epoch", or "mini-batch".
+    In each collect, :class:`Collector` collects a number of policy-env interactions, and accumulates
+    them into a replay buffer. This buffer is used as the "data" to train the policy.
+    At the end of each collect, the policy is *updated* several times.
+
+    The API has some resemblence with `PyTorch Lightning <https://pytorch-lightning.readthedocs.io/>`__,
+    but it's essentially different because this trainer is built for RL applications, and thus
+    most configurations are under RL context.
+    We are still looking for ways to incorporate existing trainer libraries, because it looks like
+    big efforts to build a trainer as powerful as those libraries, and also, that's not our primary goal.
+
+    It's essentially different
+    `tianshou's built-in trainers <https://tianshou.readthedocs.io/en/master/api/tianshou.trainer.html>`__,
+    as it's far much more complicated than that.
+
+    Parameters
+    ----------
+    max_iters
+        Maximum iterations before stopping.
+    val_every_n_iters
+        Perform validation every n iterations (i.e., training collects).
+    logger
+        Logger to record the backtest results. Logger must be present because
+        without logger, all information will be lost.
+    finite_env_type
+        Type of finite env implementation.
+    concurrency
+        Parallel workers.
+    fast_dev_run
+        Create a subset for debugging.
+        How this is implemented depends on the implementation of training vessel.
+        For :class:`~qlib.rl.vessel.TrainingVessel`, if greater than zero,
+        a random subset sized ``fast_dev_run`` will be used
+        instead of ``train_initial_states`` and ``val_initial_states``.
+    """
+
+    should_stop: bool
+    """Set to stop the training."""
+
+    metrics: dict
+    """Numeric metrics of produced in train/val/test.
+    In the middle of training / validation, metrics will be of the latest episode.
+    When each iteration of training / validation finishes, metrics will be the aggregation
+    of all episodes encountered in this iteration.
+
+    Cleared on every new iteration of training.
+
+    In fit, validation metrics will be prefixed with ``val/``.
+    """
+
+    current_iter: int
+    """Current iteration (collect) of training."""
+
+    loggers: list[LogWriter]
+    """A list of log writers."""
+
+    def __init__(
+        self,
+        *,
+        max_iters: int | None = None,
+        val_every_n_iters: int | None = None,
+        loggers: LogWriter | list[LogWriter] | None = None,
+        callbacks: list[Callback] | None = None,
+        finite_env_type: FiniteEnvType = "subproc",
+        concurrency: int = 2,
+        fast_dev_run: int | None = None,
+    ):
+        self.max_iters = max_iters
+        self.val_every_n_iters = val_every_n_iters
+
+        if isinstance(loggers, list):
+            self.loggers = loggers
+        elif isinstance(loggers, LogWriter):
+            self.loggers = [loggers]
+        else:
+            self.loggers = []
+
+        self.loggers.append(LogBuffer(self._metrics_callback, loglevel=self._min_loglevel()))
+
+        self.callbacks: list[Callback] = callbacks if callbacks is not None else []
+        self.finite_env_type = finite_env_type
+        self.concurrency = concurrency
+        self.fast_dev_run = fast_dev_run
+
+        self.current_stage: Literal["train", "val", "test"] = "train"
+
+        self.vessel: TrainingVesselBase = cast(TrainingVesselBase, None)
+
+    def initialize(self):
+        """Initialize the whole training process.
+
+        The states here should be synchronized with state_dict.
+        """
+        self.should_stop = False
+        self.current_iter = 0
+        self.current_episode = 0
+        self.current_stage = "train"
+
+    def initialize_iter(self):
+        """Initialize one iteration / collect."""
+        self.metrics = {}
+
+    def state_dict(self) -> dict:
+        """Putting every states of current training into a dict, at best effort.
+
+        It doesn't try to handle all the possible kinds of states in the middle of one training collect.
+        For most cases at the end of each iteration, things should be usually correct.
+
+        Note that it's also intended behavior that replay buffer data in the collector will be lost.
+        """
+        return {
+            "vessel": self.vessel.state_dict(),
+            "callbacks": {name: callback.state_dict() for name, callback in self.named_callbacks().items()},
+            "loggers": {name: logger.state_dict() for name, logger in self.named_loggers().items()},
+            "should_stop": self.should_stop,
+            "current_iter": self.current_iter,
+            "current_episode": self.current_episode,
+            "current_stage": self.current_stage,
+            "metrics": self.metrics,
+        }
+
+    def load_state_dict(self, state_dict: dict) -> None:
+        """Load all states into current trainer."""
+        self.vessel.load_state_dict(state_dict["vessel"])
+        for name, callback in self.named_callbacks().items():
+            callback.load_state_dict(state_dict["callbacks"][name])
+        for name, logger in self.named_loggers().items():
+            logger.load_state_dict(state_dict["loggers"][name])
+        self.should_stop = state_dict["should_stop"]
+        self.current_iter = state_dict["current_iter"]
+        self.current_episode = state_dict["current_episode"]
+        self.current_stage = state_dict["current_stage"]
+        self.metrics = state_dict["metrics"]
+
+    def named_callbacks(self) -> dict[str, Callback]:
+        """Retrieve a collection of callbacks where each one has a name.
+        Useful when saving checkpoints.
+        """
+        return _named_collection(self.callbacks)
+
+    def named_loggers(self) -> dict[str, LogWriter]:
+        """Retrieve a collection of loggers where each one has a name.
+        Useful when saving checkpoints.
+        """
+        return _named_collection(self.loggers)
+
+    def fit(self, vessel: TrainingVesselBase, ckpt_path: Path | None = None) -> None:
+        """Train the RL policy upon the defined simulator.
+
+        Parameters
+        ----------
+        vessel
+            A bundle of all elements used in training.
+        ckpt_path
+            Load a pre-trained / paused training checkpoint.
+        """
+        self.vessel = vessel
+        vessel.assign_trainer(self)
+
+        if ckpt_path is not None:
+            _logger.info("Resuming states from %s", str(ckpt_path))
+            self.load_state_dict(torch.load(ckpt_path))
+        else:
+            self.initialize()
+
+        self._call_callback_hooks("on_fit_start")
+
+        while not self.should_stop:
+            self.initialize_iter()
+
+            self._call_callback_hooks("on_iter_start")
+
+            self.current_stage = "train"
+            self._call_callback_hooks("on_train_start")
+
+            # TODO
+            # Add a feature that supports reloading the training environment every few iterations.
+            with _wrap_context(vessel.train_seed_iterator()) as iterator:
+                vector_env = self.venv_from_iterator(iterator)
+                self.vessel.train(vector_env)
+
+            self._call_callback_hooks("on_train_end")
+
+            if self.val_every_n_iters is not None and (self.current_iter + 1) % self.val_every_n_iters == 0:
+                # Implementation of validation loop
+                self.current_stage = "val"
+                self._call_callback_hooks("on_validate_start")
+                with _wrap_context(vessel.val_seed_iterator()) as iterator:
+                    vector_env = self.venv_from_iterator(iterator)
+                    self.vessel.validate(vector_env)
+
+                self._call_callback_hooks("on_validate_end")
+
+            # This iteration is considered complete.
+            # Bumping the current iteration counter.
+            self.current_iter += 1
+
+            if self.max_iters is not None and self.current_iter >= self.max_iters:
+                self.should_stop = True
+
+            self._call_callback_hooks("on_iter_end")
+
+        self._call_callback_hooks("on_fit_end")
+
+    def test(self, vessel: TrainingVesselBase) -> None:
+        """Test the RL policy against the simulator.
+
+        The simulator will be fed with data generated in ``test_seed_iterator``.
+
+        Parameters
+        ----------
+        vessel
+            A bundle of all related elements.
+        """
+        self.vessel = vessel
+        vessel.assign_trainer(self)
+
+        self.initialize_iter()
+
+        self.current_stage = "test"
+        self._call_callback_hooks("on_test_start")
+        with _wrap_context(vessel.test_seed_iterator()) as iterator:
+            vector_env = self.venv_from_iterator(iterator)
+            self.vessel.test(vector_env)
+        self._call_callback_hooks("on_test_end")
+
+    def venv_from_iterator(self, iterator: Iterable[InitialStateType]) -> FiniteVectorEnv:
+        """Create a vectorized environment from iterator and the training vessel."""
+
+        def env_factory():
+            # FIXME: state_interpreter and action_interpreter are stateful (having a weakref of env),
+            # and could be thread unsafe.
+            # I'm not sure whether it's a design flaw.
+            # I'll rethink about this when designing the trainer.
+
+            if self.finite_env_type == "dummy":
+                # We could only experience the "threading-unsafe" problem in dummy.
+                state = copy.deepcopy(self.vessel.state_interpreter)
+                action = copy.deepcopy(self.vessel.action_interpreter)
+                rew = copy.deepcopy(self.vessel.reward)
+            else:
+                state = self.vessel.state_interpreter
+                action = self.vessel.action_interpreter
+                rew = self.vessel.reward
+
+            return EnvWrapper(
+                self.vessel.simulator_fn,
+                state,
+                action,
+                iterator,
+                rew,
+                logger=LogCollector(min_loglevel=self._min_loglevel()),
+            )
+
+        return vectorize_env(
+            env_factory,
+            self.finite_env_type,
+            self.concurrency,
+            self.loggers,
+        )
+
+    def _metrics_callback(self, on_episode: bool, on_collect: bool, log_buffer: LogBuffer) -> None:
+        if on_episode:
+            # Update the global counter.
+            self.current_episode = log_buffer.global_episode
+            metrics = log_buffer.episode_metrics()
+        elif on_collect:
+            # Update the latest metrics.
+            metrics = log_buffer.collect_metrics()
+        if self.current_stage == "val":
+            metrics = {"val/" + name: value for name, value in metrics.items()}
+        self.metrics.update(metrics)
+
+    def _call_callback_hooks(self, hook_name: str, *args: Any, **kwargs: Any) -> None:
+        for callback in self.callbacks:
+            fn = getattr(callback, hook_name)
+            fn(self, self.vessel, *args, **kwargs)
+
+    def _min_loglevel(self):
+        if not self.loggers:
+            return LogLevel.PERIODIC
+        else:
+            # To save bandwidth
+            return min(lg.loglevel for lg in self.loggers)
+
+
+@contextmanager
+def _wrap_context(obj):
+    """Make any object a (possibly dummy) context manager."""
+
+    if isinstance(obj, AbstractContextManager):
+        # obj has __enter__ and __exit__
+        with obj as ctx:
+            yield ctx
+    else:
+        yield obj
+
+
+def _named_collection(seq: Sequence[T]) -> dict[str, T]:
+    """Convert a list into a dict, where each item is named with its type."""
+    res = {}
+    for item in seq:
+        typename = type(item).__name__.lower()
+        if typename not in res:
+            res[typename] = item
+        else:
+            # names are auto-labelled as earlystop1, earlystop2, ...
+            for retry in range(1, 1000):
+                if f"{typename}{retry}" not in res:
+                    res[f"{typename}{retry}"] = item
+    return res
--- a/qlib/rl/trainer/vessel.py
+++ b/qlib/rl/trainer/vessel.py
@@ -0,0 +1,214 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from __future__ import annotations
+
+import weakref
+from typing import Callable, ContextManager, Generic, Iterable, TYPE_CHECKING, Sequence, Any, TypeVar, cast, Dict
+
+import numpy as np
+from tianshou.data import Collector, VectorReplayBuffer
+from tianshou.env import BaseVectorEnv
+from tianshou.policy import BasePolicy
+
+from qlib.constant import INF
+from qlib.rl.interpreter import StateType, ActType, ObsType, PolicyActType
+from qlib.rl.simulator import InitialStateType, Simulator
+from qlib.rl.interpreter import StateInterpreter, ActionInterpreter
+from qlib.rl.reward import Reward
+from qlib.rl.utils import DataQueue
+from qlib.log import get_module_logger
+from qlib.rl.utils.finite_env import FiniteVectorEnv
+
+if TYPE_CHECKING:
+    from .trainer import Trainer
+
+
+T = TypeVar("T")
+_logger = get_module_logger(__name__)
+
+
+class SeedIteratorNotAvailable(BaseException):
+    pass
+
+
+class TrainingVesselBase(Generic[InitialStateType, StateType, ActType, ObsType, PolicyActType]):
+    """A ship that contains simulator, interpreter, and policy, will be sent to trainer.
+    This class controls algorithm-related parts of training, while trainer is responsible for runtime part.
+
+    The ship also defines the most important logic of the core training part,
+    and (optionally) some callbacks to insert customized logics at specific events.
+    """
+
+    simulator_fn: Callable[[InitialStateType], Simulator[InitialStateType, StateType, ActType]]
+    state_interpreter: StateInterpreter[StateType, ObsType]
+    action_interpreter: ActionInterpreter[StateType, PolicyActType, ActType]
+    policy: BasePolicy
+    reward: Reward
+    trainer: Trainer
+
+    def assign_trainer(self, trainer: Trainer) -> None:
+        self.trainer = weakref.proxy(trainer)  # type: ignore
+
+    def train_seed_iterator(self) -> ContextManager[Iterable[InitialStateType]] | Iterable[InitialStateType]:
+        """Override this to create a seed iterator for training.
+        If the iterable is a context manager, the whole training will be invoked in the with-block,
+        and the iterator will be automatically closed after the training is done."""
+        raise SeedIteratorNotAvailable("Seed iterator for training is not available.")
+
+    def val_seed_iterator(self) -> ContextManager[Iterable[InitialStateType]] | Iterable[InitialStateType]:
+        """Override this to create a seed iterator for validation."""
+        raise SeedIteratorNotAvailable("Seed iterator for validation is not available.")
+
+    def test_seed_iterator(self) -> ContextManager[Iterable[InitialStateType]] | Iterable[InitialStateType]:
+        """Override this to create a seed iterator for testing."""
+        raise SeedIteratorNotAvailable("Seed iterator for testing is not available.")
+
+    def train(self, vector_env: BaseVectorEnv) -> dict[str, Any]:
+        """Implement this to train one iteration. In RL, one iteration usually refers to one collect."""
+        raise NotImplementedError()
+
+    def validate(self, vector_env: FiniteVectorEnv) -> dict[str, Any]:
+        """Implement this to validate the policy once."""
+        raise NotImplementedError()
+
+    def test(self, vector_env: FiniteVectorEnv) -> dict[str, Any]:
+        """Implement this to evaluate the policy on test environment once."""
+        raise NotImplementedError()
+
+    def log(self, name: str, value: Any) -> None:
+        # FIXME: this is a workaround to make the log at least show somewhere.
+        # Need a refactor in logger to formalize this.
+        if isinstance(value, (np.ndarray, list)):
+            value = np.mean(value)
+        _logger.info(f"[Iter {self.trainer.current_iter + 1}] {name} = {value}")
+
+    def log_dict(self, data: dict[str, Any]) -> None:
+        for name, value in data.items():
+            self.log(name, value)
+
+    def state_dict(self) -> dict:
+        """Return a checkpoint of current vessel state."""
+        return {"policy": self.policy.state_dict()}
+
+    def load_state_dict(self, state_dict: dict) -> None:
+        """Restore a checkpoint from a previously saved state dict."""
+        self.policy.load_state_dict(state_dict["policy"])
+
+
+class TrainingVessel(TrainingVesselBase):
+    """The default implementation of training vessel.
+
+    ``__init__`` accepts a sequence of initial states so that iterator can be created.
+    ``train``, ``validate``, ``test`` each do one collect (and also update in train).
+    By default, the train initial states will be repeated infinitely during training,
+    and collector will control the number of episodes for each iteration.
+    In validation and testing, the val / test initial states will be used exactly once.
+
+    Extra hyper-parameters (only used in train) include:
+
+    - ``buffer_size``: Size of replay buffer.
+    - ``episode_per_iter``: Episodes per collect at training. Can be overridden by fast dev run.
+    - ``update_kwargs``: Keyword arguments appearing in ``policy.update``.
+      For example, ``dict(repeat=10, batch_size=64)``.
+    """
+
+    def __init__(
+        self,
+        *,
+        simulator_fn: Callable[[InitialStateType], Simulator[InitialStateType, StateType, ActType]],
+        state_interpreter: StateInterpreter[StateType, ObsType],
+        action_interpreter: ActionInterpreter[StateType, PolicyActType, ActType],
+        policy: BasePolicy,
+        reward: Reward,
+        train_initial_states: Sequence[InitialStateType] | None = None,
+        val_initial_states: Sequence[InitialStateType] | None = None,
+        test_initial_states: Sequence[InitialStateType] | None = None,
+        buffer_size: int = 20000,
+        episode_per_iter: int = 1000,
+        update_kwargs: dict[str, Any] = cast(Dict[str, Any], None),
+    ):
+        self.simulator_fn = simulator_fn  # type: ignore
+        self.state_interpreter = state_interpreter
+        self.action_interpreter = action_interpreter
+        self.policy = policy
+        self.reward = reward
+        self.train_initial_states = train_initial_states
+        self.val_initial_states = val_initial_states
+        self.test_initial_states = test_initial_states
+        self.buffer_size = buffer_size
+        self.episode_per_iter = episode_per_iter
+        self.update_kwargs = update_kwargs or {}
+
+    def train_seed_iterator(self) -> ContextManager[Iterable[InitialStateType]] | Iterable[InitialStateType]:
+        if self.train_initial_states is not None:
+            _logger.info("Training initial states collection size: %d", len(self.train_initial_states))
+            # Implement fast_dev_run here.
+            train_initial_states = self._random_subset("train", self.train_initial_states, self.trainer.fast_dev_run)
+            return DataQueue(train_initial_states, repeat=-1, shuffle=True)
+        return super().train_seed_iterator()
+
+    def val_seed_iterator(self) -> ContextManager[Iterable[InitialStateType]] | Iterable[InitialStateType]:
+        if self.val_initial_states is not None:
+            _logger.info("Validation initial states collection size: %d", len(self.val_initial_states))
+            val_initial_states = self._random_subset("val", self.val_initial_states, self.trainer.fast_dev_run)
+            return DataQueue(val_initial_states, repeat=1)
+        return super().val_seed_iterator()
+
+    def test_seed_iterator(self) -> ContextManager[Iterable[InitialStateType]] | Iterable[InitialStateType]:
+        if self.test_initial_states is not None:
+            _logger.info("Testing initial states collection size: %d", len(self.test_initial_states))
+            test_initial_states = self._random_subset("test", self.test_initial_states, self.trainer.fast_dev_run)
+            return DataQueue(test_initial_states, repeat=1)
+        return super().test_seed_iterator()
+
+    def train(self, vector_env: FiniteVectorEnv) -> dict[str, Any]:
+        """Create a collector and collects ``episode_per_iter`` episodes.
+        Update the policy on the collected replay buffer.
+        """
+        self.policy.train()
+
+        with vector_env.collector_guard():
+            collector = Collector(self.policy, vector_env, VectorReplayBuffer(self.buffer_size, len(vector_env)))
+
+            # Number of episodes collected in each training iteration can be overridden by fast dev run.
+            if self.trainer.fast_dev_run is not None:
+                episodes = self.trainer.fast_dev_run
+            else:
+                episodes = self.episode_per_iter
+
+            col_result = collector.collect(n_episode=episodes)
+            update_result = self.policy.update(sample_size=0, buffer=collector.buffer, **self.update_kwargs)
+            res = {**col_result, **update_result}
+            self.log_dict(res)
+            return res
+
+    def validate(self, vector_env: FiniteVectorEnv) -> dict[str, Any]:
+        self.policy.eval()
+
+        with vector_env.collector_guard():
+            test_collector = Collector(self.policy, vector_env)
+            res = test_collector.collect(n_step=INF * len(vector_env))
+            self.log_dict(res)
+            return res
+
+    def test(self, vector_env: FiniteVectorEnv) -> dict[str, Any]:
+        self.policy.eval()
+
+        with vector_env.collector_guard():
+            test_collector = Collector(self.policy, vector_env)
+            res = test_collector.collect(n_step=INF * len(vector_env))
+            self.log_dict(res)
+            return res
+
+    @staticmethod
+    def _random_subset(name: str, collection: Sequence[T], size: int | None) -> Sequence[T]:
+        if size is None:
+            # Size = None -> original collection
+            return collection
+        order = np.random.permutation(len(collection))
+        res = [collection[o] for o in order[:size]]
+        _logger.info(
+            "Fast running in development mode. Cut %s initial states from %d to %d.", name, len(collection), len(res)
+        )
+        return res
--- a/qlib/rl/utils/data_queue.py
+++ b/qlib/rl/utils/data_queue.py
@@ -145,7 +145,9 @@ class DataQueue(Generic[T]):
    def __iter__(self):
        if not self._activated:
            raise ValueError(
-                "Need to call activate() to launch a daemon worker to produce data into data queue before using it."
+                "Need to call activate() to launch a daemon worker "
+                "to produce data into data queue before using it. "
+                "You probably have forgotten to use the DataQueue in a with block."
            )
        return self._consumer()

@@ -161,19 +163,21 @@ class DataQueue(Generic[T]):
        # pytorch dataloader is used here only because we need its sampler and multi-processing
        from torch.utils.data import DataLoader, Dataset  # pylint: disable=import-outside-toplevel

-        dataloader = DataLoader(
-            cast(Dataset[T], self.dataset),
-            batch_size=None,
-            num_workers=self.producer_num_workers,
-            shuffle=self.shuffle,
-            collate_fn=lambda t: t,  # identity collate fn
-        )
-        repeat = 10**18 if self.repeat == -1 else self.repeat
-        for _rep in range(repeat):
-            for data in dataloader:
-                if self._done.value:
-                    # Already done.
-                    return
-                self._queue.put(data)
-            _logger.debug(f"Dataloader loop done. Repeat {_rep}.")
-        self.mark_as_done()
+        try:
+            dataloader = DataLoader(
+                cast(Dataset[T], self.dataset),
+                batch_size=None,
+                num_workers=self.producer_num_workers,
+                shuffle=self.shuffle,
+                collate_fn=lambda t: t,  # identity collate fn
+            )
+            repeat = 10**18 if self.repeat == -1 else self.repeat
+            for _rep in range(repeat):
+                for data in dataloader:
+                    if self._done.value:
+                        # Already done.
+                        return
+                    self._queue.put(data)
+                _logger.debug(f"Dataloader loop done. Repeat {_rep}.")
+        finally:
+            self.mark_as_done()
--- a/qlib/rl/utils/finite_env.py
+++ b/qlib/rl/utils/finite_env.py
@@ -120,12 +120,19 @@ class FiniteVectorEnv(BaseVectorEnv):
    from child workers. See :class:`qlib.rl.utils.LogWriter`.
    """

+    _logger: list[LogWriter]
+
    def __init__(
-        self, logger: LogWriter | list[LogWriter], env_fns: list[Callable[..., gym.Env]], **kwargs: Any
+        self, logger: LogWriter | list[LogWriter] | None, env_fns: list[Callable[..., gym.Env]], **kwargs: Any
    ) -> None:
        super().__init__(env_fns, **kwargs)

-        self._logger: list[LogWriter] = logger if isinstance(logger, list) else [logger]
+        if isinstance(logger, list):
+            self._logger = logger
+        elif isinstance(logger, LogWriter):
+            self._logger = [logger]
+        else:
+            self._logger = []
        self._alive_env_ids: Set[int] = set()
        self._reset_alive_envs()
        self._default_obs = self._default_info = self._default_rew = None
@@ -177,7 +184,7 @@ class FiniteVectorEnv(BaseVectorEnv):

        1. Catch and ignore the StopIteration exception, which is the stopping signal
           thrown by FiniteEnv to let tianshou know that ``collector.collect()`` should exit.
-        2. Notify the loggers that the collect is done what it's done.
+        2. Notify the loggers that the collect is ready / done what it's ready / done.

        Examples
        --------
@@ -186,6 +193,9 @@ class FiniteVectorEnv(BaseVectorEnv):
        """
        self._collector_guarded = True

+        for logger in self._logger:
+            logger.on_env_all_ready()
+
        try:
            yield self
        except StopIteration:
@@ -298,7 +308,21 @@ def vectorize_env(
    concurrency: int,
    logger: LogWriter | list[LogWriter],
 ) -> FiniteVectorEnv:
-    """Helper function to create a vector env.
+    """Helper function to create a vector env. Can be used to replace usual VectorEnv.
+
+    For example, once you wrote: ::
+
+        DummyVectorEnv([lambda: gym.make(task) for _ in range(env_num)])
+
+    Now you can replace it with: ::
+
+        finite_env_factory(lambda: gym.make(task), "dummy", env_num, my_logger)
+
+    By doing such replacement, you have two additional features enabled (compared to normal VectorEnv):
+
+    1. The vector env will check for NaN observation and kill the worker when its found.
+       See :class:`FiniteVectorEnv` for why we need this.
+    2. A logger to explicit collect logs from environment workers.

    Parameters
    ----------
--- a/qlib/rl/utils/log.py
+++ b/qlib/rl/utils/log.py
@@ -12,13 +12,16 @@ in each worker, and writes them to console, log files, or tensorboard...
 The two modules communicate by the "log" field in "info" returned by ``env.step()``.
 """

+# NOTE: This file contains many hardcoded / ad-hoc rules.
+# Refactoring it will be one of the future tasks.
+
 from __future__ import annotations

 import logging
 from collections import defaultdict
 from enum import IntEnum
 from pathlib import Path
-from typing import Any, TypeVar, Generic, Set, TYPE_CHECKING, Sequence
+from typing import Any, TypeVar, Generic, Set, TYPE_CHECKING, Sequence, Callable

 import numpy as np
 import pandas as pd
@@ -29,7 +32,7 @@ if TYPE_CHECKING:
    from .env_wrapper import InfoDict


-__all__ = ["LogCollector", "LogWriter", "LogLevel", "ConsoleWriter", "CsvWriter"]
+__all__ = ["LogCollector", "LogWriter", "LogLevel", "LogBuffer", "ConsoleWriter", "CsvWriter"]

 ObsType = TypeVar("ObsType")
 ActType = TypeVar("ActType")
@@ -175,18 +178,53 @@ class LogWriter(Generic[ObsType, ActType]):
        self.clear()

    def clear(self):
+        """Clear all the metrics for a fresh start.
+        To make the logger instance reusable.
+        """
        self.episode_count = self.step_count = 0
        self.active_env_ids = set()
-        self.logs = []

-    def aggregation(self, array: Sequence[Any]) -> Any:
+    def state_dict(self) -> dict:
+        """Save the states of the logger to a dict."""
+        return {
+            "episode_count": self.episode_count,
+            "step_count": self.step_count,
+            "global_step": self.global_step,
+            "global_episode": self.global_episode,
+            "active_env_ids": self.active_env_ids,
+            "episode_lengths": self.episode_lengths,
+            "episode_rewards": self.episode_rewards,
+            "episode_logs": self.episode_logs,
+        }
+
+    def load_state_dict(self, state_dict: dict) -> None:
+        """Load the states of current logger from a dict."""
+        self.episode_count = state_dict["episode_count"]
+        self.step_count = state_dict["step_count"]
+        self.global_step = state_dict["global_step"]
+        self.global_episode = state_dict["global_episode"]
+
+        # These are runtime infos.
+        # Though they are loaded, I don't think it really helps.
+        self.active_env_ids = state_dict["active_env_ids"]
+        self.episode_lenghts = state_dict["episode_lengths"]
+        self.episode_rewards = state_dict["episode_rewards"]
+        self.episode_logs = state_dict["episode_logs"]
+
+    def aggregation(self, array: Sequence[Any], name: str | None = None) -> Any:
        """Aggregation function from step-wise to episode-wise.

        If it's a sequence of float, take the mean.
        Otherwise, take the first element.
+
+        If a name is specified and,
+
+        - if it's ``reward``, the reduction will be sum.
        """
        assert len(array) > 0, "The aggregated array must be not empty."
        if all(isinstance(v, float) for v in array):
+            if name == "reward":
+                return np.sum(array)
            return np.mean(array)
        else:
            return array[0]
@@ -253,10 +291,93 @@ class LogWriter(Generic[ObsType, ActType]):
        self.episode_rewards[env_id] = []
        self.episode_logs[env_id] = []

+    def on_env_all_ready(self) -> None:
+        """When all environments are ready to run.
+        Usually, loggers should be reset here.
+        """
+        self.clear()
+
    def on_env_all_done(self) -> None:
        """All done. Time for cleanup."""


+class LogBuffer(LogWriter):
+    """Keep all numbers in memory.
+
+    Objects that can't be aggregated like strings, tensors, images can't be stored in the buffer.
+    To persist them, please use :class:`PickleWriter`.
+
+    Every time, Log buffer receives a new metric, the callback is triggered,
+    which is useful when tracking metrics inside a trainer.
+
+    Parameters
+    ----------
+    callback
+        A callback receiving three arguments:
+
+        - on_episode: Whether it's called at the end of an episode
+        - on_collect: Whether it's called at the end of a collect
+        - log_buffer: the :class:`LogBbuffer`object
+
+        No return value is expected.
+    """
+
+    # FIXME: needs a metric count
+
+    def __init__(self, callback: Callable[[bool, bool, LogBuffer], None], loglevel: int | LogLevel = LogLevel.PERIODIC):
+        super().__init__(loglevel)
+        self.callback = callback
+
+    def state_dict(self) -> dict:
+        return {
+            **super().state_dict(),
+            "latest_metrics": self._latest_metrics,
+            "aggregated_metrics": self._aggregated_metrics,
+        }
+
+    def load_state_dict(self, state_dict: dict) -> None:
+        self._latest_metrics = state_dict["latest_metrics"]
+        self._aggregated_metrics = state_dict["aggregated_metrics"]
+        return super().load_state_dict(state_dict)
+
+    def clear(self):
+        super().clear()
+        self._latest_metrics: dict[str, float] | None = None
+        self._aggregated_metrics: dict[str, float] = defaultdict(float)
+
+    def log_episode(self, length: int, rewards: list[float], contents: list[dict[str, Any]]) -> None:
+        # FIXME Dup of ConsoleWriter
+        episode_wise_contents: dict[str, list] = defaultdict(list)
+        for step_contents in contents:
+            for name, value in step_contents.items():
+                # FIXME This could be false-negative for some numpy types
+                if isinstance(value, float):
+                    episode_wise_contents[name].append(value)
+
+        logs: dict[str, float] = {}
+        for name, values in episode_wise_contents.items():
+            logs[name] = self.aggregation(values, name)  # type: ignore
+            self._aggregated_metrics[name] += logs[name]
+
+        self._latest_metrics = logs
+
+        self.callback(True, False, self)
+
+    def on_env_all_done(self) -> None:
+        # This happens when collect exits
+        self.callback(False, True, self)
+
+    def episode_metrics(self) -> dict[str, float]:
+        """Retrieve the numeric metrics of the latest episode."""
+        if self._latest_metrics is None:
+            raise ValueError("No episode metrics available yet.")
+        return self._latest_metrics
+
+    def collect_metrics(self) -> dict[str, float]:
+        """Retrieve the aggregated metrics of the latest collect."""
+        return {name: value / self.episode_count for name, value in self._aggregated_metrics.items()}
+
+
 class ConsoleWriter(LogWriter):
    """Write log messages to console periodically.

@@ -289,6 +410,8 @@ class ConsoleWriter(LogWriter):

        self.console_logger = get_module_logger(__name__, level=logging.INFO)

+    # FIXME: save & reload
+
    def clear(self):
        super().clear()
        # Clear average meters
@@ -308,7 +431,7 @@ class ConsoleWriter(LogWriter):
        # This should be done at every step, regardless of periodic or not.
        logs: dict[str, float] = {}
        for name, values in episode_wise_contents.items():
-            logs[name] = self.aggregation(values)  # type: ignore
+            logs[name] = self.aggregation(values, name)  # type: ignore

        for name, value in logs.items():
            self.metric_counts[name] += 1
@@ -350,6 +473,8 @@ class CsvWriter(LogWriter):

    all_records: list[dict[str, Any]]

+    # FIXME: save & reload
+
    def __init__(self, output_dir: Path, loglevel: int | LogLevel = LogLevel.PERIODIC):
        super().__init__(loglevel)
        self.output_dir = output_dir
@@ -370,7 +495,7 @@ class CsvWriter(LogWriter):

        logs: dict[str, float] = {}
        for name, values in episode_wise_contents.items():
-            logs[name] = self.aggregation(values)  # type: ignore
+            logs[name] = self.aggregation(values, name)  # type: ignore

        self.all_records.append(logs)

@@ -392,7 +517,3 @@ class TensorboardWriter(LogWriter):

 class MlflowWriter(LogWriter):
    """Add logs to mlflow."""
-
-
-class LogBuffer(LogWriter):
-    """Keep everything in memory."""
--- a/qlib/strategy/base.py
+++ b/qlib/strategy/base.py
@@ -3,7 +3,7 @@
 from __future__ import annotations

 from abc import abstractmethod
-from typing import TYPE_CHECKING, Optional
+from typing import TYPE_CHECKING, Any, Generator, Optional

 if TYPE_CHECKING:
    from qlib.backtest.exchange import Exchange
@@ -122,7 +122,10 @@ class BaseStrategy:
            self.outer_trade_decision = outer_trade_decision

    @abstractmethod
-    def generate_trade_decision(self, execute_result: list = None) -> BaseTradeDecision:
+    def generate_trade_decision(
+        self,
+        execute_result: list = None,
+    ) -> Union[BaseTradeDecision, Generator[Any, Any, BaseTradeDecision]]:
        """Generate trade decision in each trading bar

        Parameters
--- a/qlib/utils/index_data.py
+++ b/qlib/utils/index_data.py
@@ -9,6 +9,8 @@ Motivation of index_data
 `index_data` try to behave like pandas (some API will be different because we try to be simpler and more intuitive) but don't compromise the performance. It provides the basic numpy data and simple indexing feature. If users call APIs which may compromise the performance, index_data will raise Errors.
 """

+from __future__ import annotations
+
 from typing import Dict, Tuple, Union, Callable, List
 import bisect

@@ -16,7 +18,7 @@ import numpy as np
 import pandas as pd


-def concat(data_list: Union["SingleData"], axis=0) -> "MultiData":
+def concat(data_list: Union[SingleData], axis=0) -> MultiData:
    """concat all SingleData by index.
    TODO: now just for SingleData.

@@ -52,7 +54,7 @@ def concat(data_list: Union["SingleData"], axis=0) -> "MultiData":
        raise ValueError(f"axis must be 0 or 1")


-def sum_by_index(data_list: Union["SingleData"], new_index: list, fill_value=0) -> "SingleData":
+def sum_by_index(data_list: Union[SingleData], new_index: list, fill_value=0) -> SingleData:
    """concat all SingleData by new index.

    Parameters
@@ -554,7 +556,7 @@ class SingleData(IndexData):
                f"The indexes of self and other do not meet the requirements of the four arithmetic operations"
            )

-    def reindex(self, index: Index, fill_value=np.NaN):
+    def reindex(self, index: Index, fill_value=np.NaN) -> SingleData:
        """reindex data and fill the missing value with np.NaN.

        Parameters
@@ -580,7 +582,7 @@ class SingleData(IndexData):
                pass
        return SingleData(tmp_data, index)

-    def add(self, other: "SingleData", fill_value=0):
+    def add(self, other: SingleData, fill_value=0):
        # TODO: add and __add__ are a little confusing.
        # This could be a more general
        common_index = self.index | other.index
--- a/qlib/workflow/recorder.py
+++ b/qlib/workflow/recorder.py
@@ -2,11 +2,13 @@
 # Licensed under the MIT License.

 import os
+import sys
 import mlflow
 import logging
 import shutil
 import pickle
 import tempfile
+import subprocess
 from pathlib import Path
 from datetime import datetime

@@ -296,8 +298,32 @@ class MLflowRecorder(Recorder):
        # - This may cause delay when uploading results
        # - The logging time may not be accurate
        self.async_log = AsyncCaller()
+
+        # TODO: currently, this is only supported in MLflowRecorder.
+        # Maybe we can make this feature more general.
+        self._log_uncommitted_code()
+
+        self.log_params(**{"cmd-sys.argv": " ".join(sys.argv)})  # log the command to produce current experiment
        return run

+    def _log_uncommitted_code(self):
+        """
+        Mlflow only log the commit id of the current repo. But usually, user will have a lot of uncommitted changes.
+        So this tries to automatically to log them all.
+        """
+        # TODO: the sub-directories maybe git repos.
+        # So it will be better if we can walk the sub-directories and log the uncommitted changes.
+        for cmd, fname in [
+            ("git diff", "code_diff.txt"),
+            ("git status", "code_status.txt"),
+            ("git diff --cached", "code_cached.txt"),
+        ]:
+            try:
+                out = subprocess.check_output(cmd, shell=True)
+                self.client.log_text(self.id, out.decode(), fname)  # this behaves same as above
+            except subprocess.CalledProcessError:
+                logger.info(f"Fail to log the uncommitted code of $CWD when run `{cmd}`")
+
    def end_run(self, status: str = Recorder.STATUS_S):
        assert status in [
            Recorder.STATUS_S,
--- a/setup.py
+++ b/setup.py
@@ -137,10 +137,24 @@ setup(
            "sphinx",
            "sphinx_rtd_theme",
            "pre-commit",
+            # CI dependencies
+            "wheel",
+            "setuptools",
+            "black",
+            "pylint",
+            "mypy",
+            "flake8",
+            "readthedocs_sphinx_ext",
+            "cmake",
+            "lxml",
+            "baostock",
+            "yahooquery",
+            "beautifulsoup4",
+            "tianshou",
+            "gym>=0.24",  # If you do not put gym at the end, gym will degrade causing pytest results to fail.
        ],
        "rl": [
            "tianshou",
-            "gym",
            "torch",
        ],
    },
--- a/tests/pytest.ini
+++ b/tests/pytest.ini
@@ -1,4 +1,6 @@
 [pytest]
+markers =
+    slow: marks tests as slow (deselect with '-m "not slow"')
 filterwarnings =
    ignore:.*rng.randint:DeprecationWarning
    ignore:.*Casting input x to numpy array:UserWarning
--- a/tests/rl/test_logger.py
+++ b/tests/rl/test_logger.py
@@ -81,7 +81,7 @@ def test_simple_env_logger(caplog):
        line = line.strip()
        if line:
            line_counter += 1
-            assert re.match(r".*reward 42\.0000 \(42.0000\)  a .* \((4|5|6)\.\d+\)  c .* \((14|15|16)\.\d+\)", line)
+            assert re.match(r".*reward .*  a .* \((4|5|6)\.\d+\)  c .* \((14|15|16)\.\d+\)", line)
    assert line_counter >= 3


--- a/tests/rl/test_saoe_simple.py
+++ b/tests/rl/test_saoe_simple.py
@@ -17,7 +17,7 @@ from qlib.backtest import Order
 from qlib.config import C
 from qlib.log import set_log_with_config
 from qlib.rl.data import pickle_styled
-from qlib.rl.entries.test import backtest
+from qlib.rl.trainer import backtest, train
 from qlib.rl.order_execution import *
 from qlib.rl.utils import ConsoleWriter, CsvWriter, EnvWrapperStatus

@@ -306,3 +306,26 @@ def test_cn_ppo_strategy():
    assert np.isclose(metrics["pa"].mean(), -16.21578303474833)
    assert np.isclose(metrics["market_price"].mean(), 58.68277690875527)
    assert np.isclose(metrics["trade_price"].mean(), 58.76063985000002)
+
+
+def test_ppo_train():
+    set_log_with_config(C.logging_config)
+    # The data starts with 9:31 and ends with 15:00
+    orders = pickle_styled.load_orders(CN_ORDER_DIR, start_time=pd.Timestamp("9:31"), end_time=pd.Timestamp("14:58"))
+    assert len(orders) == 40
+
+    state_interp = FullHistoryStateInterpreter(CN_FEATURE_DATA_DIR, 8, 240, 6)
+    action_interp = CategoricalActionInterpreter(4)
+    network = Recurrent(state_interp.observation_space)
+    policy = PPO(network, state_interp.observation_space, action_interp.action_space, 1e-4)
+
+    train(
+        partial(SingleAssetOrderExecution, data_dir=CN_BACKTEST_DATA_DIR, ticks_per_step=30),
+        state_interp,
+        action_interp,
+        orders,
+        policy,
+        PAPenaltyReward(),
+        vessel_kwargs={"episode_per_iter": 100, "update_kwargs": {"batch_size": 64, "repeat": 5}},
+        trainer_kwargs={"max_iters": 2, "loggers": ConsoleWriter(total_episodes=100)},
+    )
--- a/tests/rl/test_trainer.py
+++ b/tests/rl/test_trainer.py
@@ -0,0 +1,202 @@
+import os
+import random
+import sys
+from pathlib import Path
+
+import pytest
+
+import torch
+import torch.nn as nn
+from gym import spaces
+from tianshou.policy import PPOPolicy
+
+from qlib.config import C
+from qlib.log import set_log_with_config
+from qlib.rl.interpreter import StateInterpreter, ActionInterpreter
+from qlib.rl.simulator import Simulator
+from qlib.rl.reward import Reward
+from qlib.rl.trainer import Trainer, TrainingVessel, EarlyStopping, Checkpoint
+
+pytestmark = pytest.mark.skipif(sys.version_info < (3, 8), reason="Pickle styled data only supports Python >= 3.8")
+
+
+class ZeroSimulator(Simulator):
+    def __init__(self, *args, **kwargs):
+        self.action = self.correct = 0
+
+    def step(self, action):
+        self.action = action
+        self.correct = action == 0
+        self._done = random.choice([False, True])
+        if self._done:
+            self.env.logger.add_scalar("acc", self.correct * 100)
+
+    def get_state(self):
+        return {
+            "acc": self.correct * 100,
+            "action": self.action,
+        }
+
+    def done(self) -> bool:
+        return self._done
+
+
+class NoopStateInterpreter(StateInterpreter):
+    observation_space = spaces.Dict(
+        {
+            "acc": spaces.Discrete(200),
+            "action": spaces.Discrete(2),
+        }
+    )
+
+    def interpret(self, simulator_state):
+        return simulator_state
+
+
+class NoopActionInterpreter(ActionInterpreter):
+    action_space = spaces.Discrete(2)
+
+    def interpret(self, simulator_state, action):
+        return action
+
+
+class AccReward(Reward):
+    def reward(self, simulator_state):
+        if self.env.status["done"]:
+            return simulator_state["acc"] / 100
+        return 0.0
+
+
+class PolicyNet(nn.Module):
+    def __init__(self, out_features=1, return_state=False):
+        super().__init__()
+        self.fc = nn.Linear(32, out_features)
+        self.return_state = return_state
+
+    def forward(self, obs, state=None, **kwargs):
+        res = self.fc(torch.randn(obs["acc"].shape[0], 32))
+        if self.return_state:
+            return nn.functional.softmax(res, dim=-1), state
+        else:
+            return res
+
+
+def _ppo_policy():
+    actor = PolicyNet(2, True)
+    critic = PolicyNet()
+    policy = PPOPolicy(
+        actor,
+        critic,
+        torch.optim.Adam(tuple(actor.parameters()) + tuple(critic.parameters())),
+        torch.distributions.Categorical,
+        action_space=NoopActionInterpreter().action_space,
+    )
+    return policy
+
+
+def test_trainer():
+    set_log_with_config(C.logging_config)
+    trainer = Trainer(max_iters=10, finite_env_type="subproc")
+    policy = _ppo_policy()
+
+    vessel = TrainingVessel(
+        simulator_fn=lambda init: ZeroSimulator(init),
+        state_interpreter=NoopStateInterpreter(),
+        action_interpreter=NoopActionInterpreter(),
+        policy=policy,
+        train_initial_states=list(range(100)),
+        val_initial_states=list(range(10)),
+        test_initial_states=list(range(10)),
+        reward=AccReward(),
+        episode_per_iter=500,
+        update_kwargs=dict(repeat=10, batch_size=64),
+    )
+    trainer.fit(vessel)
+    assert trainer.current_iter == 10
+    assert trainer.current_episode == 5000
+    assert abs(trainer.metrics["acc"] - trainer.metrics["reward"] * 100) < 1e-4
+    assert trainer.metrics["acc"] > 80
+    trainer.test(vessel)
+    assert trainer.metrics["acc"] > 60
+
+
+def test_trainer_fast_dev_run():
+    set_log_with_config(C.logging_config)
+    trainer = Trainer(max_iters=2, fast_dev_run=2, finite_env_type="shmem")
+    policy = _ppo_policy()
+
+    vessel = TrainingVessel(
+        simulator_fn=lambda init: ZeroSimulator(init),
+        state_interpreter=NoopStateInterpreter(),
+        action_interpreter=NoopActionInterpreter(),
+        policy=policy,
+        train_initial_states=list(range(100)),
+        val_initial_states=list(range(10)),
+        test_initial_states=list(range(10)),
+        reward=AccReward(),
+        episode_per_iter=500,
+        update_kwargs=dict(repeat=10, batch_size=64),
+    )
+    trainer.fit(vessel)
+    assert trainer.current_episode == 4
+
+
+def test_trainer_earlystop():
+    # TODO this is just sanity check.
+    # need to see the logs to check whether it works.
+    set_log_with_config(C.logging_config)
+    trainer = Trainer(
+        max_iters=10,
+        val_every_n_iters=1,
+        finite_env_type="dummy",
+        callbacks=[EarlyStopping("val/reward", restore_best_weights=True)],
+    )
+    policy = _ppo_policy()
+
+    vessel = TrainingVessel(
+        simulator_fn=lambda init: ZeroSimulator(init),
+        state_interpreter=NoopStateInterpreter(),
+        action_interpreter=NoopActionInterpreter(),
+        policy=policy,
+        train_initial_states=list(range(100)),
+        val_initial_states=list(range(10)),
+        test_initial_states=list(range(10)),
+        reward=AccReward(),
+        episode_per_iter=500,
+        update_kwargs=dict(repeat=10, batch_size=64),
+    )
+    trainer.fit(vessel)
+    assert trainer.metrics["val/acc"] > 30
+    assert trainer.current_iter == 2  # second iteration
+
+
+def test_trainer_checkpoint():
+    set_log_with_config(C.logging_config)
+    output_dir = Path(__file__).parent / ".output"
+    trainer = Trainer(max_iters=2, finite_env_type="dummy", callbacks=[Checkpoint(output_dir, every_n_iters=1)])
+    policy = _ppo_policy()
+
+    vessel = TrainingVessel(
+        simulator_fn=lambda init: ZeroSimulator(init),
+        state_interpreter=NoopStateInterpreter(),
+        action_interpreter=NoopActionInterpreter(),
+        policy=policy,
+        train_initial_states=list(range(100)),
+        val_initial_states=list(range(10)),
+        test_initial_states=list(range(10)),
+        reward=AccReward(),
+        episode_per_iter=100,
+        update_kwargs=dict(repeat=10, batch_size=64),
+    )
+    trainer.fit(vessel)
+
+    assert (output_dir / "001.pth").exists()
+    assert (output_dir / "002.pth").exists()
+    assert os.readlink(output_dir / "latest.pth") == str(output_dir / "002.pth")
+
+    trainer.load_state_dict(torch.load(output_dir / "001.pth"))
+    assert trainer.current_iter == 1
+    assert trainer.current_episode == 100
+
+    # Reload the checkpoint at first iteration
+    trainer.fit(vessel, ckpt_path=output_dir / "001.pth")
--- a/tests/rolling_tests/test_update_pred.py
+++ b/tests/rolling_tests/test_update_pred.py
@@ -1,5 +1,6 @@
 import copy
 import unittest
+import pytest

 import fire
 import pandas as pd
@@ -14,6 +15,7 @@ from qlib.workflow.online.update import LabelUpdater


 class TestRolling(TestAutoData):
+    @pytest.mark.slow
    def test_update_pred(self):
        """
        This test is for testing if it will raise error if the `to_date` is out of the boundary.
@@ -73,6 +75,7 @@ class TestRolling(TestAutoData):
        # this range is fixed now
        self.assertTrue((updated_pred.loc[mod_range2] == -2).all().item())

+    @pytest.mark.slow
    def test_update_label(self):

        task = copy.deepcopy(CSI300_GBDT_TASK)
--- a/tests/test_all_pipeline.py
+++ b/tests/test_all_pipeline.py
@@ -4,6 +4,7 @@
 import sys
 import shutil
 import unittest
+import pytest
 from pathlib import Path

 import qlib
@@ -184,16 +185,19 @@ class TestAllFlow(TestAutoData):
    def tearDownClass(cls) -> None:
        shutil.rmtree(cls.URI_PATH.lstrip("file:"))

+    @pytest.mark.slow
    def test_0_train_with_sigana(self):
        TestAllFlow.PRED_SCORE, ic_ric, uri_path = train_with_sigana(self.URI_PATH)
        self.assertGreaterEqual(ic_ric["ic"].all(), 0, "train failed")
        self.assertGreaterEqual(ic_ric["ric"].all(), 0, "train failed")

+    @pytest.mark.slow
    def test_1_train(self):
        TestAllFlow.PRED_SCORE, ic_ric, TestAllFlow.RID = train(self.URI_PATH)
        self.assertGreaterEqual(ic_ric["ic"].all(), 0, "train failed")
        self.assertGreaterEqual(ic_ric["ric"].all(), 0, "train failed")

+    @pytest.mark.slow
    def test_2_backtest(self):
        analyze_df = backtest_analysis(TestAllFlow.PRED_SCORE, TestAllFlow.RID, self.URI_PATH)
        self.assertGreaterEqual(
@@ -203,6 +207,7 @@ class TestAllFlow(TestAutoData):
        )
        self.assertTrue(not analyze_df.isna().any().any(), "backtest failed")

+    @pytest.mark.slow
    def test_3_expmanager(self):
        pass_default, pass_current, uri_path = fake_experiment()
        self.assertTrue(pass_default, msg="default uri is incorrect")
--- a/tests/test_contrib_workflow.py
+++ b/tests/test_contrib_workflow.py
@@ -4,6 +4,7 @@
 from qlib.workflow.record_temp import SignalRecord
 import shutil
 import unittest
+import pytest
 from pathlib import Path

 from qlib.contrib.workflow import MultiSegRecord, SignalMseRecord
@@ -47,9 +48,11 @@ class TestAllFlow(TestAutoData):
    def tearDownClass(cls) -> None:
        shutil.rmtree(cls.URI_PATH.lstrip("file:"))

+    @pytest.mark.slow
    def test_0_multiseg(self):
        uri_path = train_multiseg(self.URI_PATH)

+    @pytest.mark.slow
    def test_1_mse(self):
        uri_path = train_mse(self.URI_PATH)

--- a/tests/test_dataset.py
+++ b/tests/test_dataset.py
@@ -2,6 +2,7 @@
 # Licensed under the MIT License.

 import unittest
+import pytest
 import sys
 from qlib.tests import TestAutoData
 from qlib.data.dataset import TSDatasetH
@@ -11,6 +12,7 @@ from qlib.data.dataset.handler import DataHandlerLP


 class TestDataset(TestAutoData):
+    @pytest.mark.slow
    def testTSDataset(self):
        tsdh = TSDatasetH(
            handler={
--- a/tests/test_pit.py
+++ b/tests/test_pit.py
@@ -6,12 +6,13 @@ import sys
 import qlib
 import shutil
 import unittest
+import pytest
 import pandas as pd
 import baostock as bs
 from pathlib import Path

 from qlib.data import D
-from scripts.get_data import GetData
+from qlib.tests.data import GetData
 from scripts.dump_pit import DumpPitData

 sys.path.append(str(Path(__file__).resolve().parent.parent.joinpath("scripts/data_collector/pit")))
@@ -39,17 +40,21 @@ class TestPIT(unittest.TestCase):
        pit_dir = str(SOURCE_DIR.joinpath("pit").resolve())
        pit_normalized_dir = str(SOURCE_DIR.joinpath("pit_normalized").resolve())
        GetData().qlib_data(name="qlib_data_simple", target_dir=cn_data_dir, region="cn")
-        bs.login()
-        Run(
-            source_dir=pit_dir,
-            interval="quarterly",
-        ).download_data(start="2000-01-01", end="2020-01-01", symbol_regex="^(600519|000725).*")
+        GetData().qlib_data(name="qlib_data", target_dir=pit_dir, region="pit")
+
+        # NOTE: This code does the same thing as line 43, but since baostock is not stable in downloading data, we have chosen to download offline data.
+        # bs.login()
+        # Run(
+        #     source_dir=pit_dir,
+        #     interval="quarterly",
+        # ).download_data(start="2000-01-01", end="2020-01-01", symbol_regex="^(600519|000725).*")
+        # bs.logout()
+
        Run(
            source_dir=pit_dir,
            normalize_dir=pit_normalized_dir,
            interval="quarterly",
        ).normalize_data()
-        bs.logout()
        DumpPitData(
            csv_path=pit_normalized_dir,
            qlib_dir=cn_data_dir,
@@ -119,6 +124,7 @@ class TestPIT(unittest.TestCase):
        """
        self.check_same(data, expect)

+    @pytest.mark.slow
    def test_expr(self):
        fields = [
            "P(Mean($$roewa_q, 1))",
Author	SHA1	Message	Date
you-n-g	670ae6aa61	Update test_qlib_from_source.yml	2022-06-29 17:07:05 +08:00
you-n-g	7e3ca3c5f4	Update test_qlib_from_pip.yml	2022-06-29 17:02:57 +08:00
you-n-g	ab0174a363	Update test_qlib_from_source.yml	2022-06-29 17:01:51 +08:00
you-n-g	8c72ed99c2	Update test_qlib_from_source.yml	2022-06-29 17:00:35 +08:00
you-n-g	b9624b074f	Update test_qlib_from_source_slow.yml	2022-06-29 13:38:36 +08:00
Huoran Li	23c657a7a2	Backtest Mypy (#1130 ) * Done * Fix test errors * Revert profit_attribution.py * Minor * A minor update on collect_data type hint * Resolve PR comments * Use black to format code * Fix CI errors	2022-06-28 22:16:46 +08:00
you-n-g	9bf3423a64	Auto log uncommmitted code (#1167 ) * Auto log uncommmitted code * Support set record name & trainer; * Update recorder.py	2022-06-28 19:53:21 +08:00
Yuge Zhang	25ecb1135f	Qlib RL framework (stage 2) - trainer (#1125 ) * checkpoint (cherry picked from commit 1a8e0bd4671ee6d624a7d09bb198a273282cd050) * Not a workable version (cherry picked from commit 3498e185684cd5590d3ab97e0ab69eab8c1e0e3a) * vessel * ckpt * . * vessel * . * . * checkpoint callback * . * cleanup * logger * . * test * . * add test * . * . * . * . * New reward * Add train API * fix mypy * fix lint * More comment * 3.7 compat * fix test * fix test * . * Resolve comments * fix typehint	2022-06-28 19:53:05 +08:00
Linlang	2ca0d88d2d	change_pitdata_source (#1171 ) * change_pitdata_source * retain_normalize * add_comment	2022-06-28 16:29:59 +08:00
Linlang	50d74b5560	split_CI (#1141 )	2022-06-28 10:17:29 +08:00
you-n-g	a87b02619a	Qlib dev doc (#1142 )	2022-06-21 09:46:30 +08:00
you-n-g	da676a20a2	Add time limit for CI (#1127 ) * Add time limit for CI * Update test_macos.yml	2022-06-16 16:35:20 +08:00
you-n-g	13d904d9a9	Update Version To Dev	2022-06-15 14:53:54 +08:00