10695097128

Committed 04 Sep 2024 04:27AM UTC coverage: 83.123% (-0.2%) from 83.298%

Build # 10695097128

Build Type

push

github

Committed by

web-flow

Commit Message

Merge pull request #502 from WenjieDu/(feat)add_moderntcn

Implement ModernTCN as an imputation model

Run Details

266 of 348 new or added lines in 9 files covered. (76.44%)

1 existing line in 1 file now uncovered.

11131 of 13391 relevant lines covered (83.12%)

4.99 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

86.03

/pypots/forecasting/csdi/model.py

"""
The implementation of CSDI for the partially-observed time-series forecasting task.

"""

# Created by Wenjie Du <wenjay.du@gmail.com>
# License: BSD-3-Clause

import os
from typing import Union, Optional

import numpy as np
import torch
from torch.utils.data import DataLoader

try:
    import nni
except ImportError:
    pass

from .core import _CSDI
from .data import DatasetForForecastingCSDI, TestDatasetForForecastingCSDI
from ..base import BaseNNForecaster
from ...data.checking import key_in_data_set
from ...optim.adam import Adam
from ...optim.base import Optimizer
from ...utils.logging import logger


class CSDI(BaseNNForecaster):
    """The PyTorch implementation of the CSDI model :cite:`tashiro2021csdi`.

    Parameters
    ----------
    n_steps :
        The number of time steps in the time-series data sample.

    n_features :
        The number of features in the time-series data sample.

    n_pred_steps :
        The number of steps in the forecasting time series.

    n_pred_features :
        The number of features in the forecasting time series.

    n_layers :
        The number of layers in the CSDI model.

    n_heads :
        The number of heads in the multi-head attention mechanism.

    n_channels :
        The number of residual channels.

    d_time_embedding :
        The dimension number of the time (temporal) embedding.

    d_feature_embedding :
        The dimension number of the feature embedding.

    d_diffusion_embedding :
        The dimension number of the diffusion embedding.

    is_unconditional :
        Whether the model is unconditional or conditional.

    target_strategy :
        The strategy for selecting the target for the diffusion process. It has to be one of ["mix", "random"].

    n_diffusion_steps :
        The number of the diffusion step T in the original paper.

    schedule:
        The schedule for other noise levels. It has to be one of ["quad", "linear"].

    beta_start:
        The minimum noise level.

    beta_end:
        The maximum noise level.

    batch_size :
        The batch size for training and evaluating the model.

    epochs :
        The number of epochs for training the model.

    patience :
        The patience for the early-stopping mechanism. Given a positive integer, the training process will be
        stopped when the model does not perform better after that number of epochs.
        Leaving it default as None will disable the early-stopping.

    optimizer :
        The optimizer for model training.
        If not given, will use a default Adam optimizer.

    num_workers :
        The number of subprocesses to use for data loading.
        `0` means data loading will be in the main process, i.e. there won't be subprocesses.

    device :
        The device for the model to run on. It can be a string, a :class:`torch.device` object, or a list of them.
        If not given, will try to use CUDA devices first (will use the default CUDA device if there are multiple),
        then CPUs, considering CUDA and CPU are so far the main devices for people to train ML models.
        If given a list of devices, e.g. ['cuda:0', 'cuda:1'], or [torch.device('cuda:0'), torch.device('cuda:1')] , the
        model will be parallely trained on the multiple devices (so far only support parallel training on CUDA devices).
        Other devices like Google TPU and Apple Silicon accelerator MPS may be added in the future.

    saving_path :
        The path for automatically saving model checkpoints and tensorboard files (i.e. loss values recorded during
        training into a tensorboard file). Will not save if not given.

    model_saving_strategy :
        The strategy to save model checkpoints. It has to be one of [None, "best", "better", "all"].
        No model will be saved when it is set as None.
        The "best" strategy will only automatically save the best model after the training finished.
        The "better" strategy will automatically save the model during training whenever the model performs
        better than in previous epochs.
        The "all" strategy will save every model after each epoch training.

    verbose :
        Whether to print out the training logs during the training process.
    """

    def __init__(
        self,
        n_steps: int,
        n_features: int,
        n_pred_steps: int,
        n_pred_features: int,
        n_layers: int,
        n_heads: int,
        n_channels: int,
        d_time_embedding: int,
        d_feature_embedding: int,
        d_diffusion_embedding: int,
        n_diffusion_steps: int = 50,
        target_strategy: str = "random",
        is_unconditional: bool = False,
        schedule: str = "quad",
        beta_start: float = 0.0001,
        beta_end: float = 0.5,
        batch_size: int = 32,
        epochs: int = 100,
        patience: Optional[int] = None,
        optimizer: Optional[Optimizer] = Adam(),
        num_workers: int = 0,
        device: Optional[Union[str, torch.device, list]] = None,
        saving_path: Optional[str] = None,
        model_saving_strategy: Optional[str] = "best",
        verbose: bool = True,
    ):
        super().__init__(
            batch_size,
            epochs,
            patience,
            num_workers,
            device,
            saving_path,
            model_saving_strategy,
            verbose,
        )
        assert n_pred_features == n_features, (
            f"currently n_pred_features of CSDI forecasting model should be equal to n_features, "
            f"but got {n_pred_features} and {n_features}."
        )
        assert target_strategy in ["mix", "random"]
        assert schedule in ["quad", "linear"]
        self.n_steps = n_steps
        self.n_features = n_features
        self.n_pred_steps = n_pred_steps
        self.n_pred_features = n_pred_features
        self.target_strategy = target_strategy

        # set up the model
        self.model = _CSDI(
            n_features,
            n_pred_features,
            n_layers,
            n_heads,
            n_channels,
            d_time_embedding,
            d_feature_embedding,
            d_diffusion_embedding,
            is_unconditional,
            n_diffusion_steps,
            schedule,
            beta_start,
            beta_end,
        )
        self._print_model_size()
        self._send_model_to_given_device()

        # set up the optimizer
        self.optimizer = optimizer
        self.optimizer.init_optimizer(self.model.parameters())

    def _assemble_input_for_training(self, data: list) -> dict:
        (
            indices,
            X_ori,
            indicating_mask,
            cond_mask,
            observed_tp,
            feature_id,
        ) = self._send_data_to_given_device(data)

        inputs = {
            "X_ori": X_ori.permute(0, 2, 1),  # ori observed part for model hint
            "indicating_mask": indicating_mask.permute(0, 2, 1),  # for loss calc
            "cond_mask": cond_mask.permute(0, 2, 1),  # for masking X_ori
            "observed_tp": observed_tp,
            "feature_id": feature_id,
        }
        return inputs

    def _assemble_input_for_validating(self, data: list) -> dict:
        return self._assemble_input_for_training(data)

    def _assemble_input_for_testing(self, data: list) -> dict:
        (
            indices,
            X,
            cond_mask,
            observed_tp,
            feature_id,
        ) = self._send_data_to_given_device(data)

        inputs = {
            "X": X.permute(0, 2, 1),  # for model input
            "cond_mask": cond_mask.permute(0, 2, 1),  # missing mask
            "observed_tp": observed_tp,
            "feature_id": feature_id,
        }
        return inputs

    def _train_model(
        self,
        training_loader: DataLoader,
        val_loader: DataLoader = None,
    ) -> None:
        # each training starts from the very beginning, so reset the loss and model dict here
        self.best_loss = float("inf")
        self.best_model_dict = None

        try:
            training_step = 0
            for epoch in range(1, self.epochs + 1):
                self.model.train()
                epoch_train_loss_collector = []
                for idx, data in enumerate(training_loader):
                    training_step += 1
                    inputs = self._assemble_input_for_training(data)
                    self.optimizer.zero_grad()
                    results = self.model.forward(inputs)
                    # use sum() before backward() in case of multi-gpu training
                    results["loss"].sum().backward()
                    self.optimizer.step()
                    epoch_train_loss_collector.append(results["loss"].sum().item())

                    # save training loss logs into the tensorboard file for every step if in need
                    if self.summary_writer is not None:
                        self._save_log_into_tb_file(training_step, "training", results)

                # mean training loss of the current epoch
                mean_train_loss = np.mean(epoch_train_loss_collector)

                if val_loader is not None:
                    self.model.eval()
                    val_loss_collector = []
                    with torch.no_grad():
                        for idx, data in enumerate(val_loader):
                            inputs = self._assemble_input_for_validating(data)
                            results = self.model.forward(
                                inputs, training=False, n_sampling_times=0
                            )
                            val_loss_collector.append(results["loss"].sum().item())

                    mean_val_loss = np.asarray(val_loss_collector).mean()

                    # save validation loss logs into the tensorboard file for every epoch if in need
                    if self.summary_writer is not None:
                        val_loss_dict = {
                            "validating_loss": mean_val_loss,
                        }
                        self._save_log_into_tb_file(epoch, "validating", val_loss_dict)

                    logger.info(
                        f"Epoch {epoch:03d} - "
                        f"training loss: {mean_train_loss:.4f}, "
                        f"validation loss: {mean_val_loss:.4f}"
                    )
                    mean_loss = mean_val_loss
                else:
                    logger.info(
                        f"Epoch {epoch:03d} - training loss: {mean_train_loss:.4f}"
                    )
                    mean_loss = mean_train_loss

                if np.isnan(mean_loss):
                    logger.warning(
                        f"‼️ Attention: got NaN loss in Epoch {epoch}. This may lead to unexpected errors."
                    )

                if mean_loss < self.best_loss:
                    self.best_epoch = epoch
                    self.best_loss = mean_loss
                    self.best_model_dict = self.model.state_dict()
                    self.patience = self.original_patience
                else:
                    self.patience -= 1

                # save the model if necessary
                self._auto_save_model_if_necessary(
                    confirm_saving=self.best_epoch == epoch,
                    saving_name=f"{self.__class__.__name__}_epoch{epoch}_loss{mean_loss}",
                )

                if os.getenv("enable_tuning", False):
                    nni.report_intermediate_result(mean_loss)
                    if epoch == self.epochs - 1 or self.patience == 0:
                        nni.report_final_result(self.best_loss)

                if self.patience == 0:
                    logger.info(
                        "Exceeded the training patience. Terminating the training procedure..."
                    )
                    break

        except KeyboardInterrupt:  # if keyboard interrupt, only warning
            logger.warning("‼️ Training got interrupted by the user. Exist now ...")
        except Exception as e:  # other kind of exception follows below processing
            logger.error(f"❌ Exception: {e}")
            if self.best_model_dict is None:  # if no best model, raise error
                raise RuntimeError(
                    "Training got interrupted. Model was not trained. Please investigate the error printed above."
                )
            else:
                RuntimeWarning(
                    "Training got interrupted. Please investigate the error printed above.\n"
                    "Model got trained and will load the best checkpoint so far for testing.\n"
                    "If you don't want it, please try fit() again."
                )

        if np.isnan(self.best_loss):
            raise ValueError("Something is wrong. best_loss is Nan after training.")

        logger.info(
            f"Finished training. The best model is from epoch#{self.best_epoch}."
        )

    def fit(
        self,
        train_set: Union[dict, str],
        val_set: Optional[Union[dict, str]] = None,
        file_type: str = "hdf5",
        n_sampling_times: int = 1,
    ) -> None:
        # Step 1: wrap the input data with classes Dataset and DataLoader
        training_set = DatasetForForecastingCSDI(
            train_set,
            file_type=file_type,
        )
        training_loader = DataLoader(
            training_set,
            batch_size=self.batch_size,
            shuffle=True,
            num_workers=self.num_workers,
        )
        val_loader = None
        if val_set is not None:
            if not key_in_data_set("X_pred", val_set):
                raise ValueError("val_set must contain 'X_pred' for model validation.")
            val_set = DatasetForForecastingCSDI(
                val_set,
                file_type=file_type,
            )
            val_loader = DataLoader(
                val_set,
                batch_size=self.batch_size,
                shuffle=False,
                num_workers=self.num_workers,
            )

        # Step 2: train the model and freeze it
        self._train_model(training_loader, val_loader)
        self.model.load_state_dict(self.best_model_dict)
        self.model.eval()  # set the model as eval status to freeze it.

        # Step 3: save the model if necessary
        self._auto_save_model_if_necessary(confirm_saving=True)

    def predict(
        self,
        test_set: Union[dict, str],
        file_type: str = "hdf5",
        n_sampling_times: int = 1,
    ) -> dict:
        """

        Parameters
        ----------
        test_set : dict or str
            The dataset for model validating, should be a dictionary including keys as 'X' and 'y',
            or a path string locating a data file.
            If it is a dict, X should be array-like of shape [n_samples, sequence length (n_steps), n_features],
            which is time-series data for validating, can contain missing values, and y should be array-like of shape
            [n_samples], which is classification labels of X.
            If it is a path string, the path should point to a data file, e.g. a h5 file, which contains
            key-value pairs like a dict, and it has to include keys as 'X' and 'y'.

        file_type :
            The type of the given file if test_set is a path string.

        n_sampling_times:
            The number of sampling times for the model to sample from the diffusion process.

        Returns
        -------
        result_dict: dict
            Prediction results in a Python Dictionary for the given samples.
            It should be a dictionary including a key named 'imputation'.

        """
        assert n_sampling_times > 0, "n_sampling_times should be greater than 0."

        # Step 1: wrap the input data with classes Dataset and DataLoader
        self.model.eval()  # set the model as eval status to freeze it.
        test_set = TestDatasetForForecastingCSDI(
            test_set,
            self.n_pred_steps,
            self.n_pred_features,
            file_type=file_type,
        )
        test_loader = DataLoader(
            test_set,
            batch_size=self.batch_size,
            shuffle=False,
            num_workers=self.num_workers,
        )
        forecasting_collector = []

        # Step 2: process the data with the model
        with torch.no_grad():
            for idx, data in enumerate(test_loader):
                inputs = self._assemble_input_for_testing(data)
                results = self.model(
                    inputs,
                    training=False,
                    n_sampling_times=n_sampling_times,
                )
                forecasting_data = results["forecasting_data"][
                    :, :, -self.n_pred_steps :
                ]
                forecasting_collector.append(forecasting_data)

        # Step 3: output collection and return
        forecasting_data = torch.cat(forecasting_collector).cpu().detach().numpy()
        result_dict = {
            "forecasting": forecasting_data,  # [bz, n_sampling_times, n_pred_steps, n_features]
        }
        return result_dict

    def forecast(
        self,
        test_set: Union[dict, str],
        file_type: str = "hdf5",
    ) -> np.ndarray:
        """Forecast the future of the input with the trained model.

        Parameters
        ----------
        test_set :
            The data samples for testing, should be array-like of shape [n_samples, sequence length (n_steps),
            n_features], or a path string locating a data file, e.g. h5 file.

        file_type :
            The type of the given file if X is a path string.

        Returns
        -------
        array-like, shape [n_samples, n_pred_steps, n_features],
            Forecasting results.
        """

        result_dict = self.predict(test_set, file_type=file_type)
        return result_dict["forecasting"]

1	"""	6✔
2	The implementation of CSDI for the partially-observed time-series forecasting task.
3
4	"""
5
6	# Created by Wenjie Du <wenjay.du@gmail.com>
7	# License: BSD-3-Clause
8
9	import os	6✔
10	from typing import Union, Optional	6✔
11
12	import numpy as np	6✔
13	import torch	6✔
14	from torch.utils.data import DataLoader	6✔
15
16	try:	6✔
17	import nni	6✔
18	except ImportError:	6✔
19	pass	6✔
20
21	from .core import _CSDI	6✔
22	from .data import DatasetForForecastingCSDI, TestDatasetForForecastingCSDI	6✔
23	from ..base import BaseNNForecaster	6✔
24	from ...data.checking import key_in_data_set	6✔
25	from ...optim.adam import Adam	6✔
26	from ...optim.base import Optimizer	6✔
27	from ...utils.logging import logger	6✔
28
29
30	class CSDI(BaseNNForecaster):	6✔
31	"""The PyTorch implementation of the CSDI model :cite:`tashiro2021csdi`.	6✔
32
33	Parameters
34	----------
35	n_steps :
36	The number of time steps in the time-series data sample.
37
38	n_features :
39	The number of features in the time-series data sample.
40
41	n_pred_steps :
42	The number of steps in the forecasting time series.
43
44	n_pred_features :
45	The number of features in the forecasting time series.
46
47	n_layers :
48	The number of layers in the CSDI model.
49
50	n_heads :
51	The number of heads in the multi-head attention mechanism.
52
53	n_channels :
54	The number of residual channels.
55
56	d_time_embedding :
57	The dimension number of the time (temporal) embedding.
58
59	d_feature_embedding :
60	The dimension number of the feature embedding.
61
62	d_diffusion_embedding :
63	The dimension number of the diffusion embedding.
64
65	is_unconditional :
66	Whether the model is unconditional or conditional.
67
68	target_strategy :
69	The strategy for selecting the target for the diffusion process. It has to be one of ["mix", "random"].
70
71	n_diffusion_steps :
72	The number of the diffusion step T in the original paper.
73
74	schedule:
75	The schedule for other noise levels. It has to be one of ["quad", "linear"].
76
77	beta_start:
78	The minimum noise level.
79
80	beta_end:
81	The maximum noise level.
82
83	batch_size :
84	The batch size for training and evaluating the model.
85
86	epochs :
87	The number of epochs for training the model.
88
89	patience :
90	The patience for the early-stopping mechanism. Given a positive integer, the training process will be
91	stopped when the model does not perform better after that number of epochs.
92	Leaving it default as None will disable the early-stopping.
93
94	optimizer :
95	The optimizer for model training.
96	If not given, will use a default Adam optimizer.
97
98	num_workers :
99	The number of subprocesses to use for data loading.
100	`0` means data loading will be in the main process, i.e. there won't be subprocesses.
101
102	device :
103	The device for the model to run on. It can be a string, a :class:`torch.device` object, or a list of them.
104	If not given, will try to use CUDA devices first (will use the default CUDA device if there are multiple),
105	then CPUs, considering CUDA and CPU are so far the main devices for people to train ML models.
106	If given a list of devices, e.g. ['cuda:0', 'cuda:1'], or [torch.device('cuda:0'), torch.device('cuda:1')] , the
107	model will be parallely trained on the multiple devices (so far only support parallel training on CUDA devices).
108	Other devices like Google TPU and Apple Silicon accelerator MPS may be added in the future.
109
110	saving_path :
111	The path for automatically saving model checkpoints and tensorboard files (i.e. loss values recorded during
112	training into a tensorboard file). Will not save if not given.
113
114	model_saving_strategy :
115	The strategy to save model checkpoints. It has to be one of [None, "best", "better", "all"].
116	No model will be saved when it is set as None.
117	The "best" strategy will only automatically save the best model after the training finished.
118	The "better" strategy will automatically save the model during training whenever the model performs
119	better than in previous epochs.
120	The "all" strategy will save every model after each epoch training.
121
122	verbose :
123	Whether to print out the training logs during the training process.
124	"""
125
126	def __init__(	6✔
127	self,
128	n_steps: int,
129	n_features: int,
130	n_pred_steps: int,
131	n_pred_features: int,
132	n_layers: int,
133	n_heads: int,
134	n_channels: int,
135	d_time_embedding: int,
136	d_feature_embedding: int,
137	d_diffusion_embedding: int,
138	n_diffusion_steps: int = 50,
139	target_strategy: str = "random",
140	is_unconditional: bool = False,
141	schedule: str = "quad",
142	beta_start: float = 0.0001,
143	beta_end: float = 0.5,
144	batch_size: int = 32,
145	epochs: int = 100,
146	patience: Optional[int] = None,
147	optimizer: Optional[Optimizer] = Adam(),
148	num_workers: int = 0,
149	device: Optional[Union[str, torch.device, list]] = None,
150	saving_path: Optional[str] = None,
151	model_saving_strategy: Optional[str] = "best",
152	verbose: bool = True,
153	):
154	super().__init__(	6✔
155	batch_size,
156	epochs,
157	patience,
158	num_workers,
159	device,
160	saving_path,
161	model_saving_strategy,
162	verbose,
163	)
164	assert n_pred_features == n_features, (	6✔
165	f"currently n_pred_features of CSDI forecasting model should be equal to n_features, "
166	f"but got {n_pred_features} and {n_features}."
167	)
168	assert target_strategy in ["mix", "random"]	6✔
169	assert schedule in ["quad", "linear"]	6✔
170	self.n_steps = n_steps	6✔
171	self.n_features = n_features	6✔
172	self.n_pred_steps = n_pred_steps	6✔
173	self.n_pred_features = n_pred_features	6✔
174	self.target_strategy = target_strategy	6✔
175
176	# set up the model
177	self.model = _CSDI(	6✔
178	n_features,
179	n_pred_features,
180	n_layers,
181	n_heads,
182	n_channels,
183	d_time_embedding,
184	d_feature_embedding,
185	d_diffusion_embedding,
186	is_unconditional,
187	n_diffusion_steps,
188	schedule,
189	beta_start,
190	beta_end,
191	)
192	self._print_model_size()	6✔
193	self._send_model_to_given_device()	6✔
194
195	# set up the optimizer
196	self.optimizer = optimizer	6✔
197	self.optimizer.init_optimizer(self.model.parameters())	6✔
198
199	def _assemble_input_for_training(self, data: list) -> dict:	6✔
200	(	6✔
201	indices,
202	X_ori,
203	indicating_mask,
204	cond_mask,
205	observed_tp,
206	feature_id,
207	) = self._send_data_to_given_device(data)
208
209	inputs = {	6✔
210	"X_ori": X_ori.permute(0, 2, 1), # ori observed part for model hint
211	"indicating_mask": indicating_mask.permute(0, 2, 1), # for loss calc
212	"cond_mask": cond_mask.permute(0, 2, 1), # for masking X_ori
213	"observed_tp": observed_tp,
214	"feature_id": feature_id,
215	}
216	return inputs	6✔
217
218	def _assemble_input_for_validating(self, data: list) -> dict:	6✔
219	return self._assemble_input_for_training(data)	6✔
220
221	def _assemble_input_for_testing(self, data: list) -> dict:	6✔
222	(	6✔
223	indices,
224	X,
225	cond_mask,
226	observed_tp,
227	feature_id,
228	) = self._send_data_to_given_device(data)
229
230	inputs = {	6✔
231	"X": X.permute(0, 2, 1), # for model input
232	"cond_mask": cond_mask.permute(0, 2, 1), # missing mask
233	"observed_tp": observed_tp,
234	"feature_id": feature_id,
235	}
236	return inputs	6✔
237
238	def _train_model(	6✔
239	self,
240	training_loader: DataLoader,
241	val_loader: DataLoader = None,
242	) -> None:
243	# each training starts from the very beginning, so reset the loss and model dict here
244	self.best_loss = float("inf")	6✔
245	self.best_model_dict = None	6✔
246
247	try:	6✔
248	training_step = 0	6✔
249	for epoch in range(1, self.epochs + 1):	6✔
250	self.model.train()	6✔
251	epoch_train_loss_collector = []	6✔
252	for idx, data in enumerate(training_loader):	6✔
253	training_step += 1	6✔
254	inputs = self._assemble_input_for_training(data)	6✔
255	self.optimizer.zero_grad()	6✔
256	results = self.model.forward(inputs)	6✔
257	# use sum() before backward() in case of multi-gpu training
258	results["loss"].sum().backward()	6✔
259	self.optimizer.step()	6✔
260	epoch_train_loss_collector.append(results["loss"].sum().item())	6✔
261
262	# save training loss logs into the tensorboard file for every step if in need
263	if self.summary_writer is not None:	6✔
264	self._save_log_into_tb_file(training_step, "training", results)	6✔
265
266	# mean training loss of the current epoch
267	mean_train_loss = np.mean(epoch_train_loss_collector)	6✔
268
269	if val_loader is not None:	6✔
270	self.model.eval()	6✔
271	val_loss_collector = []	6✔
272	with torch.no_grad():	6✔
273	for idx, data in enumerate(val_loader):	6✔
274	inputs = self._assemble_input_for_validating(data)	6✔
275	results = self.model.forward(	6✔
276	inputs, training=False, n_sampling_times=0
277	)
278	val_loss_collector.append(results["loss"].sum().item())	6✔
279
280	mean_val_loss = np.asarray(val_loss_collector).mean()	6✔
281
282	# save validation loss logs into the tensorboard file for every epoch if in need
283	if self.summary_writer is not None:	6✔
284	val_loss_dict = {	6✔
285	"validating_loss": mean_val_loss,
286	}
287	self._save_log_into_tb_file(epoch, "validating", val_loss_dict)	6✔
288
289	logger.info(	6✔
290	f"Epoch {epoch:03d} - "
291	f"training loss: {mean_train_loss:.4f}, "
292	f"validation loss: {mean_val_loss:.4f}"
293	)
294	mean_loss = mean_val_loss	6✔
295	else:
296	logger.info(	×
297	f"Epoch {epoch:03d} - training loss: {mean_train_loss:.4f}"
298	)
299	mean_loss = mean_train_loss	×
300
301	if np.isnan(mean_loss):	6✔
302	logger.warning(	×
303	f"‼️ Attention: got NaN loss in Epoch {epoch}. This may lead to unexpected errors."
304	)
305
306	if mean_loss < self.best_loss:	6✔
307	self.best_epoch = epoch	6✔
308	self.best_loss = mean_loss	6✔
309	self.best_model_dict = self.model.state_dict()	6✔
310	self.patience = self.original_patience	6✔
311	else:
UNCOV 312	self.patience -= 1	4✔
313
314	# save the model if necessary
315	self._auto_save_model_if_necessary(	6✔
316	confirm_saving=self.best_epoch == epoch,
317	saving_name=f"{self.__class__.__name__}_epoch{epoch}_loss{mean_loss}",
318	)
319
320	if os.getenv("enable_tuning", False):	6✔
321	nni.report_intermediate_result(mean_loss)	×
322	if epoch == self.epochs - 1 or self.patience == 0:	×
323	nni.report_final_result(self.best_loss)	×
324
325	if self.patience == 0:	6✔
326	logger.info(	×
327	"Exceeded the training patience. Terminating the training procedure..."
328	)
329	break	×
330
331	except KeyboardInterrupt: # if keyboard interrupt, only warning	×
332	logger.warning("‼️ Training got interrupted by the user. Exist now ...")	×
333	except Exception as e: # other kind of exception follows below processing	×
334	logger.error(f"❌ Exception: {e}")	×
335	if self.best_model_dict is None: # if no best model, raise error	×
336	raise RuntimeError(	×
337	"Training got interrupted. Model was not trained. Please investigate the error printed above."
338	)
339	else:
340	RuntimeWarning(	×
341	"Training got interrupted. Please investigate the error printed above.\n"
342	"Model got trained and will load the best checkpoint so far for testing.\n"
343	"If you don't want it, please try fit() again."
344	)
345
346	if np.isnan(self.best_loss):	6✔
347	raise ValueError("Something is wrong. best_loss is Nan after training.")	×
348
349	logger.info(	6✔
350	f"Finished training. The best model is from epoch#{self.best_epoch}."
351	)
352
353	def fit(	6✔
354	self,
355	train_set: Union[dict, str],
356	val_set: Optional[Union[dict, str]] = None,
357	file_type: str = "hdf5",
358	n_sampling_times: int = 1,
359	) -> None:
360	# Step 1: wrap the input data with classes Dataset and DataLoader
361	training_set = DatasetForForecastingCSDI(	6✔
362	train_set,
363	file_type=file_type,
364	)
365	training_loader = DataLoader(	6✔
366	training_set,
367	batch_size=self.batch_size,
368	shuffle=True,
369	num_workers=self.num_workers,
370	)
371	val_loader = None	6✔
372	if val_set is not None:	6✔
373	if not key_in_data_set("X_pred", val_set):	6✔
374	raise ValueError("val_set must contain 'X_pred' for model validation.")	×
375	val_set = DatasetForForecastingCSDI(	6✔
376	val_set,
377	file_type=file_type,
378	)
379	val_loader = DataLoader(	6✔
380	val_set,
381	batch_size=self.batch_size,
382	shuffle=False,
383	num_workers=self.num_workers,
384	)
385
386	# Step 2: train the model and freeze it
387	self._train_model(training_loader, val_loader)	6✔
388	self.model.load_state_dict(self.best_model_dict)	6✔
389	self.model.eval() # set the model as eval status to freeze it.	6✔
390
391	# Step 3: save the model if necessary
392	self._auto_save_model_if_necessary(confirm_saving=True)	6✔
393
394	def predict(	6✔
395	self,
396	test_set: Union[dict, str],
397	file_type: str = "hdf5",
398	n_sampling_times: int = 1,
399	) -> dict:
400	"""
401
402	Parameters
403	----------
404	test_set : dict or str
405	The dataset for model validating, should be a dictionary including keys as 'X' and 'y',
406	or a path string locating a data file.
407	If it is a dict, X should be array-like of shape [n_samples, sequence length (n_steps), n_features],
408	which is time-series data for validating, can contain missing values, and y should be array-like of shape
409	[n_samples], which is classification labels of X.
410	If it is a path string, the path should point to a data file, e.g. a h5 file, which contains
411	key-value pairs like a dict, and it has to include keys as 'X' and 'y'.
412
413	file_type :
414	The type of the given file if test_set is a path string.
415
416	n_sampling_times:
417	The number of sampling times for the model to sample from the diffusion process.
418
419	Returns
420	-------
421	result_dict: dict
422	Prediction results in a Python Dictionary for the given samples.
423	It should be a dictionary including a key named 'imputation'.
424
425	"""
426	assert n_sampling_times > 0, "n_sampling_times should be greater than 0."	6✔
427
428	# Step 1: wrap the input data with classes Dataset and DataLoader
429	self.model.eval() # set the model as eval status to freeze it.	6✔
430	test_set = TestDatasetForForecastingCSDI(	6✔
431	test_set,
432	self.n_pred_steps,
433	self.n_pred_features,
434	file_type=file_type,
435	)
436	test_loader = DataLoader(	6✔
437	test_set,
438	batch_size=self.batch_size,
439	shuffle=False,
440	num_workers=self.num_workers,
441	)
442	forecasting_collector = []	6✔
443
444	# Step 2: process the data with the model
445	with torch.no_grad():	6✔
446	for idx, data in enumerate(test_loader):	6✔
447	inputs = self._assemble_input_for_testing(data)	6✔
448	results = self.model(	6✔
449	inputs,
450	training=False,
451	n_sampling_times=n_sampling_times,
452	)
453	forecasting_data = results["forecasting_data"][	6✔
454	:, :, -self.n_pred_steps :
455	]
456	forecasting_collector.append(forecasting_data)	6✔
457
458	# Step 3: output collection and return
459	forecasting_data = torch.cat(forecasting_collector).cpu().detach().numpy()	6✔
460	result_dict = {	6✔
461	"forecasting": forecasting_data, # [bz, n_sampling_times, n_pred_steps, n_features]
462	}
463	return result_dict	6✔
464
465	def forecast(	6✔
466	self,
467	test_set: Union[dict, str],
468	file_type: str = "hdf5",
469	) -> np.ndarray:
470	"""Forecast the future of the input with the trained model.
471
472	Parameters
473	----------
474	test_set :
475	The data samples for testing, should be array-like of shape [n_samples, sequence length (n_steps),
476	n_features], or a path string locating a data file, e.g. h5 file.
477
478	file_type :
479	The type of the given file if X is a path string.
480
481	Returns
482	-------
483	array-like, shape [n_samples, n_pred_steps, n_features],
484	Forecasting results.
485	"""
486
487	result_dict = self.predict(test_set, file_type=file_type)	×
488	return result_dict["forecasting"]	×

WenjieDu / PyPOTS / 10695097128

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous