Enhance pytorch nn (#917)

* enhance pytorch_nn

* fix dim bug

* Black format

* Fix pylint error

qlib/contrib/meta/data_selection/utils.py

@@ -9,6 +9,9 @@ from torch import nn
 class ICLoss(nn.Module):
     def forward(self, pred, y, idx, skip_size=50):
         """forward.
+        FIXME:
+        - Some times it will be a slightly different from the result from `pandas.corr()`
+        - It may be caused by the precision problem of model;
 
         :param pred:
         :param y:

qlib/contrib/model/gbdt.py

@@ -10,6 +10,7 @@ from ...data.dataset import DatasetH
 from ...data.dataset.handler import DataHandlerLP
 from ...model.interpret.base import LightGBMFInt
 from ...data.dataset.weight import Reweighter
+from qlib.workflow import R
 
 
 class LGBModel(ModelFT, LightGBMFInt):
@@ -59,10 +60,12 @@ class LGBModel(ModelFT, LightGBMFInt):
         num_boost_round=None,
         early_stopping_rounds=None,
         verbose_eval=20,
-        evals_result=dict(),
+        evals_result=None,
         reweighter=None,
-        **kwargs
+        **kwargs,
     ):
+        if evals_result is None:
+            evals_result = {}  # in case of unsafety of Python default values
         ds_l = self._prepare_data(dataset, reweighter)
         ds, names = list(zip(*ds_l))
         self.model = lgb.train(
@@ -76,10 +79,13 @@ class LGBModel(ModelFT, LightGBMFInt):
             ),
             verbose_eval=verbose_eval,
             evals_result=evals_result,
-            **kwargs
+            **kwargs,
         )
         for k in names:
-            evals_result[k] = list(evals_result[k].values())[0]
+            for key, val in evals_result[k].items():
+                name = f"{key}.{k}"
+                for epoch, m in enumerate(val):
+                    R.log_metrics(**{name.replace("@", "_"): m}, step=epoch)
 
     def predict(self, dataset: DatasetH, segment: Union[Text, slice] = "test"):
         if self.model is None:

qlib/contrib/model/pytorch_gats.py

@@ -263,8 +263,8 @@ class GATs(Model):
 
         model_dict = self.GAT_model.state_dict()
         pretrained_dict = {
-            k: v for k, v in pretrained_model.state_dict().items() if k in model_dict
-        }  # pylint: disable=E1135
+            k: v for k, v in pretrained_model.state_dict().items() if k in model_dict  # pylint: disable=E1135
+        }
         model_dict.update(pretrained_dict)
         self.GAT_model.load_state_dict(model_dict)
         self.logger.info("Loading pretrained model Done...")

qlib/contrib/model/pytorch_gats_ts.py

@@ -278,8 +278,8 @@ class GATs(Model):
 
         model_dict = self.GAT_model.state_dict()
         pretrained_dict = {
-            k: v for k, v in pretrained_model.state_dict().items() if k in model_dict
-        }  # pylint: disable=E1135
+            k: v for k, v in pretrained_model.state_dict().items() if k in model_dict  # pylint: disable=E1135
+        }
         model_dict.update(pretrained_dict)
         self.GAT_model.load_state_dict(model_dict)
         self.logger.info("Loading pretrained model Done...")

qlib/contrib/model/pytorch_nn.py

@@ -4,11 +4,12 @@
 
 from __future__ import division
 from __future__ import print_function
+from collections import defaultdict
 
 import os
 import numpy as np
 import pandas as pd
-from typing import Text, Union
+from typing import Callable, Optional, Text, Union
 from sklearn.metrics import roc_auc_score, mean_squared_error
 
 import torch
@@ -20,9 +21,18 @@ from ...model.base import Model
 from ...data.dataset import DatasetH
 from ...data.dataset.handler import DataHandlerLP
 from ...data.dataset.weight import Reweighter
-from ...utils import unpack_archive_with_buffer, save_multiple_parts_file, get_or_create_path
+from ...utils import (
+    auto_filter_kwargs,
+    init_instance_by_config,
+    unpack_archive_with_buffer,
+    save_multiple_parts_file,
+    get_or_create_path,
+)
 from ...log import get_module_logger
 from ...workflow import R
+from qlib.contrib.meta.data_selection.utils import ICLoss
+from torch.nn import DataParallel
+from torch.utils.data import DataLoader, SequentialSampler
 
 
 class DNNModelPytorch(Model):
@@ -49,9 +59,6 @@ class DNNModelPytorch(Model):
 
     def __init__(
         self,
-        input_dim=360,
-        output_dim=1,
-        layers=(256,),
         lr=0.001,
         max_steps=300,
         batch_size=2000,
@@ -64,14 +71,23 @@ class DNNModelPytorch(Model):
         GPU=0,
         seed=None,
         weight_decay=0.0,
-        **kwargs
+        data_parall=False,
+        scheduler: Optional[Union[Callable]] = "default",  # when it is Callable, it accept one argument named optimizer
+        init_model=None,
+        eval_train_metric=True,
+        pt_model_uri="qlib.contrib.model.pytorch_nn.Net",
+        pt_model_kwargs={
+            "input_dim": 360,
+            "layers": (256,),
+        },
+        valid_key=DataHandlerLP.DK_L,
+        # TODO: Infer Key is a more reasonable key. But it requires more detailed processing on label processing
     ):
         # Set logger.
         self.logger = get_module_logger("DNNModelPytorch")
         self.logger.info("DNN pytorch version...")
 
         # set hyper-parameters.
-        self.layers = layers
         self.lr = lr
         self.max_steps = max_steps
         self.batch_size = batch_size
@@ -81,41 +97,36 @@ class DNNModelPytorch(Model):
         self.lr_decay_steps = lr_decay_steps
         self.optimizer = optimizer.lower()
         self.loss_type = loss
-        self.device = torch.device("cuda:%d" % (GPU) if torch.cuda.is_available() and GPU >= 0 else "cpu")
+        if isinstance(GPU, str):
+            self.device = torch.device(GPU)
+        else:
+            self.device = torch.device("cuda:%d" % (GPU) if torch.cuda.is_available() and GPU >= 0 else "cpu")
         self.seed = seed
         self.weight_decay = weight_decay
+        self.data_parall = data_parall
+        self.eval_train_metric = eval_train_metric
+        self.valid_key = valid_key
+
+        self.best_step = None
 
         self.logger.info(
             "DNN parameters setting:"
-            "\nlayers : {}"
-            "\nlr : {}"
-            "\nmax_steps : {}"
-            "\nbatch_size : {}"
-            "\nearly_stop_rounds : {}"
-            "\neval_steps : {}"
-            "\nlr_decay : {}"
-            "\nlr_decay_steps : {}"
-            "\noptimizer : {}"
-            "\nloss_type : {}"
-            "\nseed : {}"
-            "\ndevice : {}"
-            "\nuse_GPU : {}"
-            "\nweight_decay : {}".format(
-                layers,
-                lr,
-                max_steps,
-                batch_size,
-                early_stop_rounds,
-                eval_steps,
-                lr_decay,
-                lr_decay_steps,
-                optimizer,
-                loss,
-                seed,
-                self.device,
-                self.use_gpu,
-                weight_decay,
-            )
+            f"\nlr : {lr}"
+            f"\nmax_steps : {max_steps}"
+            f"\nbatch_size : {batch_size}"
+            f"\nearly_stop_rounds : {early_stop_rounds}"
+            f"\neval_steps : {eval_steps}"
+            f"\nlr_decay : {lr_decay}"
+            f"\nlr_decay_steps : {lr_decay_steps}"
+            f"\noptimizer : {optimizer}"
+            f"\nloss_type : {loss}"
+            f"\nseed : {seed}"
+            f"\ndevice : {self.device}"
+            f"\nuse_GPU : {self.use_gpu}"
+            f"\nweight_decay : {weight_decay}"
+            f"\nenable data parall : {self.data_parall}"
+            f"\npt_model_uri: {pt_model_uri}"
+            f"\npt_model_kwargs: {pt_model_kwargs}"
         )
 
         if self.seed is not None:
@@ -126,7 +137,14 @@ class DNNModelPytorch(Model):
             raise NotImplementedError("loss {} is not supported!".format(loss))
         self._scorer = mean_squared_error if loss == "mse" else roc_auc_score
 
-        self.dnn_model = Net(input_dim, output_dim, layers, loss=self.loss_type)
+        if init_model is None:
+            self.dnn_model = init_instance_by_config({"class": pt_model_uri, "kwargs": pt_model_kwargs})
+
+            if self.data_parall:
+                self.dnn_model = DataParallel(self.dnn_model).to(self.device)
+        else:
+            self.dnn_model = init_model
+
         self.logger.info("model:\n{:}".format(self.dnn_model))
         self.logger.info("model size: {:.4f} MB".format(count_parameters(self.dnn_model)))
 
@@ -137,19 +155,24 @@ class DNNModelPytorch(Model):
         else:
             raise NotImplementedError("optimizer {} is not supported!".format(optimizer))
 
-        # Reduce learning rate when loss has stopped decrease
-        self.scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
-            self.train_optimizer,
-            mode="min",
-            factor=0.5,
-            patience=10,
-            verbose=True,
-            threshold=0.0001,
-            threshold_mode="rel",
-            cooldown=0,
-            min_lr=0.00001,
-            eps=1e-08,
-        )
+        if scheduler == "default":
+            # Reduce learning rate when loss has stopped decrease
+            self.scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
+                self.train_optimizer,
+                mode="min",
+                factor=0.5,
+                patience=10,
+                verbose=True,
+                threshold=0.0001,
+                threshold_mode="rel",
+                cooldown=0,
+                min_lr=0.00001,
+                eps=1e-08,
+            )
+        elif scheduler is None:
+            self.scheduler = None
+        else:
+            self.scheduler = scheduler(optimizer=self.train_optimizer)
 
         self.fitted = False
         self.dnn_model.to(self.device)
@@ -166,40 +189,44 @@ class DNNModelPytorch(Model):
         save_path=None,
         reweighter=None,
     ):
-        df_train, df_valid = dataset.prepare(
-            ["train", "valid"], col_set=["feature", "label"], data_key=DataHandlerLP.DK_L
-        )
-        x_train, y_train = df_train["feature"], df_train["label"]
-        x_valid, y_valid = df_valid["feature"], df_valid["label"]
+        has_valid = "valid" in dataset.segments
+        segments = ["train", "valid"]
+        vars = ["x", "y", "w"]
+        all_df = defaultdict(dict)  # x_train, x_valid y_train, y_valid w_train, w_valid
+        all_t = defaultdict(dict)  # tensors
+        for seg in segments:
+            if seg in dataset.segments:
+                # df_train df_valid
+                df = dataset.prepare(
+                    seg, col_set=["feature", "label"], data_key=self.valid_key if seg == "valid" else DataHandlerLP.DK_L
+                )
+                all_df["x"][seg] = df["feature"]
+                all_df["y"][seg] = df["label"]
+                if reweighter is None:
+                    all_df["w"][seg] = pd.DataFrame(np.ones_like(all_df["y"][seg].values), index=df.index)
+                elif isinstance(reweighter, Reweighter):
+                    all_df["w"][seg] = pd.DataFrame(reweighter.reweight(df))
+                else:
+                    raise ValueError("Unsupported reweighter type.")
 
-        if reweighter is None:
-            w_train = pd.DataFrame(np.ones_like(y_train.values), index=y_train.index)
-            w_valid = pd.DataFrame(np.ones_like(y_valid.values), index=y_valid.index)
-        elif isinstance(reweighter, Reweighter):
-            w_train = pd.DataFrame(reweighter.reweight(df_train))
-            w_valid = pd.DataFrame(reweighter.reweight(df_valid))
-        else:
-            raise ValueError("Unsupported reweighter type.")
+                # get tensors
+                for v in vars:
+                    all_t[v][seg] = torch.from_numpy(all_df[v][seg].values).float()
+                    # if seg == "valid": # accelerate the eval of validation
+                    all_t[v][seg] = all_t[v][seg].to(self.device)  # This will consume a lot of memory !!!!
+
+                evals_result[seg] = []
 
         save_path = get_or_create_path(save_path)
         stop_steps = 0
         train_loss = 0
         best_loss = np.inf
-        evals_result["train"] = []
-        evals_result["valid"] = []
         # train
         self.logger.info("training...")
         self.fitted = True
         # return
         # prepare training data
-        x_train_values = torch.from_numpy(x_train.values).float()
-        y_train_values = torch.from_numpy(y_train.values).float()
-        w_train_values = torch.from_numpy(w_train.values).float()
-        train_num = y_train_values.shape[0]
-        # prepare validation data
-        x_val_auto = torch.from_numpy(x_valid.values).float().to(self.device)
-        y_val_auto = torch.from_numpy(y_valid.values).float().to(self.device)
-        w_val_auto = torch.from_numpy(w_valid.values).float().to(self.device)
+        train_num = all_t["y"]["train"].shape[0]
 
         for step in range(1, self.max_steps + 1):
             if stop_steps >= self.early_stop_rounds:
@@ -210,9 +237,9 @@ class DNNModelPytorch(Model):
             self.dnn_model.train()
             self.train_optimizer.zero_grad()
             choice = np.random.choice(train_num, self.batch_size)
-            x_batch_auto = x_train_values[choice].to(self.device)
-            y_batch_auto = y_train_values[choice].to(self.device)
-            w_batch_auto = w_train_values[choice].to(self.device)
+            x_batch_auto = all_t["x"]["train"][choice].to(self.device)
+            y_batch_auto = all_t["y"]["train"][choice].to(self.device)
+            w_batch_auto = all_t["w"]["train"][choice].to(self.device)
 
             # forward
             preds = self.dnn_model(x_batch_auto)
@@ -226,44 +253,84 @@ class DNNModelPytorch(Model):
             train_loss += loss.val
             # for evert `eval_steps` steps or at the last steps, we will evaluate the model.
             if step % self.eval_steps == 0 or step == self.max_steps:
-                stop_steps += 1
-                train_loss /= self.eval_steps
+                if has_valid:
+                    stop_steps += 1
+                    train_loss /= self.eval_steps
 
-                with torch.no_grad():
-                    self.dnn_model.eval()
-                    loss_val = AverageMeter()
+                    with torch.no_grad():
+                        self.dnn_model.eval()
 
-                    # forward
-                    preds = self.dnn_model(x_val_auto)
-                    cur_loss_val = self.get_loss(preds, w_val_auto, y_val_auto, self.loss_type)
-                    loss_val.update(cur_loss_val.item())
-                R.log_metrics(val_loss=loss_val.val, step=step)
-                if verbose:
-                    self.logger.info(
-                        "[Step {}]: train_loss {:.6f}, valid_loss {:.6f}".format(step, train_loss, loss_val.val)
-                    )
-                evals_result["train"].append(train_loss)
-                evals_result["valid"].append(loss_val.val)
-                if loss_val.val < best_loss:
+                        # forward
+                        preds = self._nn_predict(all_t["x"]["valid"], return_cpu=False)
+                        cur_loss_val = self.get_loss(preds, all_t["w"]["valid"], all_t["y"]["valid"], self.loss_type)
+                        loss_val = cur_loss_val.item()
+                        metric_val = (
+                            self.get_metric(
+                                preds.reshape(-1), all_t["y"]["valid"].reshape(-1), all_df["x"]["valid"].index
+                            )
+                            .detach()
+                            .cpu()
+                            .numpy()
+                            .item()
+                        )
+                        R.log_metrics(val_loss=loss_val, step=step)
+                        R.log_metrics(val_metric=metric_val, step=step)
+
+                        if self.eval_train_metric:
+                            metric_train = (
+                                self.get_metric(
+                                    self._nn_predict(all_t["x"]["train"], return_cpu=False),
+                                    all_t["y"]["train"].reshape(-1),
+                                    all_df["x"]["train"].index,
+                                )
+                                .detach()
+                                .cpu()
+                                .numpy()
+                                .item()
+                            )
+                            R.log_metrics(train_metric=metric_train, step=step)
+                        else:
+                            metric_train = -1
                     if verbose:
                         self.logger.info(
-                            "\tvalid loss update from {:.6f} to {:.6f}, save checkpoint.".format(
-                                best_loss, loss_val.val
-                            )
+                            f"[Step {step}]: train_loss {train_loss:.6f}, valid_loss {loss_val:.6f}, train_metric {metric_train:.6f}, valid_metric {metric_val:.6f}"
                         )
-                    best_loss = loss_val.val
-                    stop_steps = 0
-                    torch.save(self.dnn_model.state_dict(), save_path)
-                train_loss = 0
-                # update learning rate
-                self.scheduler.step(cur_loss_val)
+                    evals_result["train"].append(train_loss)
+                    evals_result["valid"].append(loss_val)
+                    if loss_val < best_loss:
+                        if verbose:
+                            self.logger.info(
+                                "\tvalid loss update from {:.6f} to {:.6f}, save checkpoint.".format(
+                                    best_loss, loss_val
+                                )
+                            )
+                        best_loss = loss_val
+                        self.best_step = step
+                        R.log_metrics(best_step=self.best_step, step=step)
+                        stop_steps = 0
+                        torch.save(self.dnn_model.state_dict(), save_path)
+                    train_loss = 0
+                    # update learning rate
+                    if self.scheduler is not None:
+                        auto_filter_kwargs(self.scheduler.step, warning=False)(metrics=cur_loss_val, epoch=step)
+                    R.log_metrics(lr=self.get_lr(), step=step)
+                else:
+                    # retraining mode
+                    if self.scheduler is not None:
+                        self.scheduler.step(epoch=step)
 
-        # restore the optimal parameters after training
-        self.dnn_model.load_state_dict(torch.load(save_path, map_location=self.device))
+        if has_valid:
+            # restore the optimal parameters after training
+            self.dnn_model.load_state_dict(torch.load(save_path, map_location=self.device))
         if self.use_gpu:
             torch.cuda.empty_cache()
 
+    def get_lr(self):
+        assert len(self.train_optimizer.param_groups) == 1
+        return self.train_optimizer.param_groups[0]["lr"]
+
     def get_loss(self, pred, w, target, loss_type):
+        pred, w, target = pred.reshape(-1), w.reshape(-1), target.reshape(-1)
         if loss_type == "mse":
             sqr_loss = torch.mul(pred - target, pred - target)
             loss = torch.mul(sqr_loss, w).mean()
@@ -274,15 +341,54 @@ class DNNModelPytorch(Model):
         else:
             raise NotImplementedError("loss {} is not supported!".format(loss_type))
 
+    def get_metric(self, pred, target, index):
+        # NOTE: the order of the index must follow <datetime, instrument> sorted order
+        return -ICLoss()(pred, target, index)  # pylint: disable=E1130
+
+    def _nn_predict(self, data, return_cpu=True):
+        """Reusing predicting NN.
+        Scenarios
+        1) test inference (data may come from CPU and expect the output data is on CPU)
+        2) evaluation on training (data may come from GPU)
+        """
+        if isinstance(data, torch.Tensor) and data.device.type != "cpu":
+            # GPU data
+            # CUDA data don't support pin_memory and multi-processing workers
+            num_workers = 0
+            pin_memory = False
+        else:
+            # CPU data
+            if not isinstance(data, torch.Tensor):
+                if isinstance(data, pd.DataFrame):
+                    data = data.values
+            # else: CPU Tensor
+            num_workers = 8
+            pin_memory = True
+        data_loader = DataLoader(
+            data,
+            sampler=SequentialSampler(data),
+            batch_size=self.batch_size,
+            drop_last=False,
+            num_workers=num_workers,
+            pin_memory=pin_memory,
+        )
+        preds = []
+        self.dnn_model.eval()
+        with torch.no_grad():
+            for x in data_loader:
+                preds.append(self.dnn_model(x.to(self.device)).detach().reshape(-1))
+        if return_cpu:
+            preds = np.concatenate([pr.cpu().numpy() for pr in preds])
+        else:
+            preds = torch.cat(preds, axis=0)
+        return preds
+
     def predict(self, dataset: DatasetH, segment: Union[Text, slice] = "test"):
         if not self.fitted:
             raise ValueError("model is not fitted yet!")
         x_test_pd = dataset.prepare(segment, col_set="feature", data_key=DataHandlerLP.DK_I)
-        x_test = torch.from_numpy(x_test_pd.values).float().to(self.device)
-        self.dnn_model.eval()
-        with torch.no_grad():
-            preds = self.dnn_model(x_test).detach().cpu().numpy()
-        return pd.Series(np.squeeze(preds), index=x_test_pd.index)
+        preds = self._nn_predict(x_test_pd)
+        return pd.Series(preds.reshape(-1), index=x_test_pd.index)
 
     def save(self, filename, **kwargs):
         with save_multiple_parts_file(filename) as model_dir:
@@ -322,16 +428,22 @@ class AverageMeter:
 
 
 class Net(nn.Module):
-    def __init__(self, input_dim, output_dim, layers=(256, 512, 768, 512, 256, 128, 64), loss="mse"):
+    def __init__(self, input_dim, output_dim=1, layers=(256,), act="LeakyReLU"):
         super(Net, self).__init__()
+
         layers = [input_dim] + list(layers)
         dnn_layers = []
         drop_input = nn.Dropout(0.05)
         dnn_layers.append(drop_input)
-        hidden_units = None
+        hidden_units = input_dim
         for i, (_input_dim, hidden_units) in enumerate(zip(layers[:-1], layers[1:])):
             fc = nn.Linear(_input_dim, hidden_units)
-            activation = nn.LeakyReLU(negative_slope=0.1, inplace=False)
+            if act == "LeakyReLU":
+                activation = nn.LeakyReLU(negative_slope=0.1, inplace=False)
+            elif act == "SiLU":
+                activation = nn.SiLU()
+            else:
+                raise NotImplementedError(f"This type of input is not supported")
             bn = nn.BatchNorm1d(hidden_units)
             seq = nn.Sequential(fc, bn, activation)
             dnn_layers.append(seq)