pypots/imputation/gpvae/model.py
"""
The implementation of GP-VAE for the partially-observed time-series imputation task.
"""
# Created by Jun Wang <jwangfx@connect.ust.hk> and 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 _GPVAE
from .data import DatasetForGPVAE
from ..base import BaseNNImputer
from ...data.checking import key_in_data_set
from ...optim.adam import Adam
from ...optim.base import Optimizer
from ...utils.logging import logger
from ...utils.metrics import calc_mse
class GPVAE(BaseNNImputer):
"""The PyTorch implementation of the GPVAE model :cite:`fortuin2020gpvae`.
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.
latent_size : int,
The feature dimension of the latent embedding
encoder_sizes : tuple,
The tuple of the network size in encoder
decoder_sizes : tuple,
The tuple of the network size in decoder
beta : float,
The weight of KL divergence in ELBO.
M : int,
The number of Monte Carlo samples for ELBO estimation during training.
K : int,
The number of importance weights for IWAE model training loss.
kernel: str
The type of kernel function chosen in the Gaussain Process Proir. ["cauchy", "diffusion", "rbf", "matern"]
sigma : float,
The scale parameter for a kernel function
length_scale : float,
The length scale parameter for a kernel function
kernel_scales : int,
The number of different length scales over latent space dimensions
window_size : int,
Window size for the inference CNN.
batch_size : int
The batch size for training and evaluating the model.
epochs : int
The number of epochs for training the model.
patience : int
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 : pypots.optim.base.Optimizer
The optimizer for model training.
If not given, will use a default Adam optimizer.
num_workers : int
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 : :class:`torch.device` or list
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 : str
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 : str
The strategy to save model checkpoints. It has to be one of [None, "best", "better"].
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.
"""
def __init__(
self,
n_steps: int,
n_features: int,
latent_size: int,
encoder_sizes: tuple = (64, 64),
decoder_sizes: tuple = (64, 64),
kernel: str = "cauchy",
beta: float = 0.2,
M: int = 1,
K: int = 1,
sigma: float = 1.0,
length_scale: float = 7.0,
kernel_scales: int = 1,
window_size: int = 3,
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: str = None,
model_saving_strategy: Optional[str] = "best",
):
super().__init__(
batch_size,
epochs,
patience,
num_workers,
device,
saving_path,
model_saving_strategy,
)
available_kernel_type = ["cauchy", "diffusion", "rbf", "matern"]
assert (
kernel in available_kernel_type
), f"kernel should be one of {available_kernel_type}, but got {kernel}"
self.n_steps = n_steps
self.n_features = n_features
self.latent_size = latent_size
self.kernel = kernel
self.encoder_sizes = encoder_sizes
self.decoder_sizes = decoder_sizes
self.beta = beta
self.M = M
self.K = K
self.sigma = sigma
self.length_scale = length_scale
self.kernel_scales = kernel_scales
# set up the model
self.model = _GPVAE(
input_dim=self.n_features,
time_length=self.n_steps,
latent_dim=self.latent_size,
kernel=self.kernel,
encoder_sizes=self.encoder_sizes,
decoder_sizes=self.decoder_sizes,
beta=self.beta,
M=self.M,
K=self.K,
sigma=self.sigma,
length_scale=self.length_scale,
kernel_scales=self.kernel_scales,
window_size=window_size,
)
self._send_model_to_given_device()
self._print_model_size()
# set up the optimizer
self.optimizer = optimizer
self.optimizer.init_optimizer(self.model.parameters())
def _assemble_input_for_training(self, data: list) -> dict:
# fetch data
(
indices,
X,
missing_mask,
) = self._send_data_to_given_device(data)
# assemble input data
inputs = {
"indices": indices,
"X": X,
"missing_mask": missing_mask,
}
return inputs
def _assemble_input_for_validating(self, data: list) -> dict:
# fetch data
(
indices,
X,
missing_mask,
X_ori,
indicating_mask,
) = self._send_data_to_given_device(data)
# assemble input data
inputs = {
"indices": indices,
"X": X,
"missing_mask": missing_mask,
"X_ori": X_ori,
"indicating_mask": indicating_mask,
}
return inputs
def _assemble_input_for_testing(self, data: list) -> dict:
return self._assemble_input_for_training(data)
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()
imputation_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=1
)
imputed_data = results["imputed_data"].mean(axis=1)
imputation_mse = (
calc_mse(
imputed_data,
inputs["X_ori"],
inputs["indicating_mask"],
)
.sum()
.detach()
.item()
)
imputation_loss_collector.append(imputation_mse)
mean_val_loss = np.mean(imputation_loss_collector)
# save validation loss logs into the tensorboard file for every epoch if in need
if self.summary_writer is not None:
val_loss_dict = {
"imputation_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=mean_loss < self.best_loss,
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",
) -> None:
# Step 1: wrap the input data with classes Dataset and DataLoader
training_set = DatasetForGPVAE(
train_set, return_X_ori=False, return_y=False, 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_ori", val_set):
raise ValueError("val_set must contain 'X_ori' for model validation.")
val_set = DatasetForGPVAE(
val_set, return_X_ori=True, return_y=False, 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 produce predictions.
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."
self.model.eval() # set the model as eval status to freeze it.
test_set = DatasetForGPVAE(
test_set, return_X_ori=False, return_y=False, file_type=file_type
)
test_loader = DataLoader(
test_set,
batch_size=self.batch_size,
shuffle=False,
num_workers=self.num_workers,
)
imputation_collector = []
with torch.no_grad():
for idx, data in enumerate(test_loader):
inputs = self._assemble_input_for_testing(data)
results = self.model.forward(
inputs, training=False, n_sampling_times=n_sampling_times
)
imputed_data = results["imputed_data"]
imputation_collector.append(imputed_data)
imputation = torch.cat(imputation_collector).cpu().detach().numpy()
result_dict = {
"imputation": imputation,
}
return result_dict
def impute(
self,
test_set: Union[dict, str],
file_type: str = "hdf5",
) -> np.ndarray:
"""Impute missing values in the given data 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, sequence length (n_steps), n_features],
Imputed data.
"""
results_dict = self.predict(test_set, file_type=file_type)
return results_dict["imputation"]