liuyu/qlib
1
0
Fork 0
mirror of https://github.com/microsoft/qlib.git synced 2026-07-10 14:26:56 +08:00
Code Activity

Migrate backtest logic from NT (#1263)

Browse Source 
* Backtest migration

* Minor bug fix in test

* Reorganize file to avoid loop import

* Fix test SAOE bug

* Remove unnecessary names

* Resolve PR comments; remove private classes;

* Fix CI error

* Resolve PR comments

* Refactor data interfaces

* Remove convert_instance_config and change config

* Pylint issue

* Pylint issue

* Fix tempfile warning

* Resolve PR comments

* Add more comments
This commit is contained in:
Huoran Li
2022-09-19 14:54:26 +08:00
committed by GitHub
parent e762548295
commit bee05f56ef
19 changed files with 794 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tests/rl/test_qlib_simulator.py
View File

@@ -11,6 +11,7 @@ from qlib.backtest.decision import Order, OrderDir, TradeRangeByTime
from qlib.backtest.executor import SimulatorExecutor
from qlib.rl.order_execution import CategoricalActionInterpreter
from qlib.rl.order_execution.simulator_qlib import SingleAssetOrderExecution
from qlib.rl.utils.env_wrapper import CollectDataEnvWrapper
TOTAL_POSITION = 2100.0
@@ -192,6 +193,8 @@ def test_interpreter() -> None:
order = get_order()
simulator = get_simulator(order)
interpreter_action = CategoricalActionInterpreter(values=NUM_EXECUTION)
interpreter_action.env = CollectDataEnvWrapper()
interpreter_action.env.reset()
NUM_STEPS = 7
state = simulator.get_state()

47
tests/rl/test_saoe_simple.py
View File

@@ -16,9 +16,11 @@ 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.data.pickle_styled import PickleProcessedDataProvider
from qlib.rl.order_execution import *
from qlib.rl.trainer import backtest, train
from qlib.rl.utils import ConsoleWriter, CsvWriter, EnvWrapperStatus
from qlib.rl.utils.env_wrapper import CollectDataEnvWrapper
pytestmark = pytest.mark.skipif(sys.version_info < (3, 8), reason="Pickle styled data only supports Python >= 3.8")
@@ -40,16 +42,15 @@ def test_pickle_data_inspect():
data = pickle_styled.load_simple_intraday_backtest_data(BACKTEST_DATA_DIR, "AAL", "2013-12-11", "close", 0)
assert len(data) == 390
data = pickle_styled.load_intraday_processed_data(
DATA_DIR / "processed", "AAL", "2013-12-11", 5, data.get_time_index()
)
provider = PickleProcessedDataProvider(DATA_DIR / "processed")
data = provider.get_data("AAL", "2013-12-11", 5, data.get_time_index())
assert len(data.today) == len(data.yesterday) == 390
def test_simulator_first_step():
order = Order("AAL", 30.0, 0, pd.Timestamp("2013-12-11 00:00:00"), pd.Timestamp("2013-12-11 23:59:59"))
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
state = simulator.get_state()
assert state.cur_time == pd.Timestamp("2013-12-11 09:30:00")
assert state.position == 30.0
@@ -83,7 +84,7 @@ def test_simulator_first_step():
def test_simulator_stop_twap():
order = Order("AAL", 13.0, 0, pd.Timestamp("2013-12-11 00:00:00"), pd.Timestamp("2013-12-11 23:59:59"))
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
for _ in range(13):
simulator.step(1.0)
@@ -106,10 +107,10 @@ def test_simulator_stop_early():
order = Order("AAL", 1.0, 1, pd.Timestamp("2013-12-11 00:00:00"), pd.Timestamp("2013-12-11 23:59:59"))
with pytest.raises(ValueError):
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
simulator.step(2.0)
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
simulator.step(1.0)
with pytest.raises(AssertionError):
@@ -119,7 +120,7 @@ def test_simulator_stop_early():
def test_simulator_start_middle():
order = Order("AAL", 15.0, 1, pd.Timestamp("2013-12-11 10:15:00"), pd.Timestamp("2013-12-11 15:44:59"))
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
assert len(simulator.ticks_for_order) == 330
assert simulator.cur_time == pd.Timestamp("2013-12-11 10:15:00")
simulator.step(2.0)
@@ -138,7 +139,7 @@ def test_simulator_start_middle():
def test_interpreter():
order = Order("AAL", 15.0, 1, pd.Timestamp("2013-12-11 10:15:00"), pd.Timestamp("2013-12-11 15:44:59"))
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
assert len(simulator.ticks_for_order) == 330
assert simulator.cur_time == pd.Timestamp("2013-12-11 10:15:00")
@@ -146,7 +147,7 @@ def test_interpreter():
class EmulateEnvWrapper(NamedTuple):
status: EnvWrapperStatus
interpreter = FullHistoryStateInterpreter(FEATURE_DATA_DIR, 13, 390, 5)
interpreter = FullHistoryStateInterpreter(13, 390, 5, PickleProcessedDataProvider(FEATURE_DATA_DIR))
interpreter_step = CurrentStepStateInterpreter(13)
interpreter_action = CategoricalActionInterpreter(20)
interpreter_action_twap = TwapRelativeActionInterpreter()
@@ -185,6 +186,10 @@ def test_interpreter():
assert np.sum(obs["data_processed"][60:]) == 0
# second step: action
interpreter_action.env = CollectDataEnvWrapper()
interpreter_action_twap.env = CollectDataEnvWrapper()
interpreter_action.env.reset()
interpreter_action_twap.env.reset()
action = interpreter_action(simulator.get_state(), 1)
assert action == 15 / 20
@@ -219,13 +224,13 @@ def test_network_sanity():
# we won't check the correctness of networks here
order = Order("AAL", 15.0, 1, pd.Timestamp("2013-12-11 9:30:00"), pd.Timestamp("2013-12-11 15:59:59"))
simulator = SingleAssetOrderExecution(order, BACKTEST_DATA_DIR)
simulator = SingleAssetOrderExecutionSimple(order, BACKTEST_DATA_DIR)
assert len(simulator.ticks_for_order) == 390
class EmulateEnvWrapper(NamedTuple):
status: EnvWrapperStatus
interpreter = FullHistoryStateInterpreter(FEATURE_DATA_DIR, 13, 390, 5)
interpreter = FullHistoryStateInterpreter(13, 390, 5, PickleProcessedDataProvider(FEATURE_DATA_DIR))
action_interp = CategoricalActionInterpreter(13)
wrapper_status_kwargs = dict(initial_state=order, obs_history=[], action_history=[], reward_history=[])
@@ -253,13 +258,15 @@ def test_twap_strategy(finite_env_type):
orders = pickle_styled.load_orders(ORDER_DIR)
assert len(orders) == 248
state_interp = FullHistoryStateInterpreter(FEATURE_DATA_DIR, 13, 390, 5)
state_interp = FullHistoryStateInterpreter(13, 390, 5, PickleProcessedDataProvider(FEATURE_DATA_DIR))
action_interp = TwapRelativeActionInterpreter()
action_interp.env = CollectDataEnvWrapper()
action_interp.env.reset()
policy = AllOne(state_interp.observation_space, action_interp.action_space)
csv_writer = CsvWriter(Path(__file__).parent / ".output")
backtest(
partial(SingleAssetOrderExecution, data_dir=BACKTEST_DATA_DIR, ticks_per_step=30),
partial(SingleAssetOrderExecutionSimple, data_dir=BACKTEST_DATA_DIR, ticks_per_step=30),
state_interp,
action_interp,
orders,
@@ -282,15 +289,17 @@ def test_cn_ppo_strategy():
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)
state_interp = FullHistoryStateInterpreter(8, 240, 6, PickleProcessedDataProvider(CN_FEATURE_DATA_DIR))
action_interp = CategoricalActionInterpreter(4)
action_interp.env = CollectDataEnvWrapper()
action_interp.env.reset()
network = Recurrent(state_interp.observation_space)
policy = PPO(network, state_interp.observation_space, action_interp.action_space, 1e-4)
policy.load_state_dict(torch.load(CN_POLICY_WEIGHTS_DIR / "ppo_recurrent_30min.pth", map_location="cpu"))
csv_writer = CsvWriter(Path(__file__).parent / ".output")
backtest(
partial(SingleAssetOrderExecution, data_dir=CN_BACKTEST_DATA_DIR, ticks_per_step=30),
partial(SingleAssetOrderExecutionSimple, data_dir=CN_BACKTEST_DATA_DIR, ticks_per_step=30),
state_interp,
action_interp,
orders,
@@ -313,13 +322,15 @@ def test_ppo_train():
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)
state_interp = FullHistoryStateInterpreter(8, 240, 6, PickleProcessedDataProvider(CN_FEATURE_DATA_DIR))
action_interp = CategoricalActionInterpreter(4)
action_interp.env = CollectDataEnvWrapper()
action_interp.env.reset()
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),
partial(SingleAssetOrderExecutionSimple, data_dir=CN_BACKTEST_DATA_DIR, ticks_per_step=30),
state_interp,
action_interp,
orders,