Source code for pypots.anomaly_detection.etsformer.model
"""
The implementation of ETSformer for the partially-observed time-series anomaly detection task.
"""
# Created by Yiyuan Yang <yyy1997sjz@gmail.com>
# License: BSD-3-Clause
from typing import Union, Optional
import torch
from torch.utils.data import DataLoader
from ..base import BaseNNDetector
from ...data.checking import key_in_data_set
from ...imputation.etsformer.core import _ETSformer
from ...imputation.saits.data import DatasetForSAITS
from ...nn.modules.loss import Criterion, MAE, MSE
from ...optim.adam import Adam
from ...optim.base import Optimizer
[docs]
class ETSformer(BaseNNDetector):
"""The PyTorch implementation of the ETSformer model for the anomaly detection task.
ETSformer is originally proposed by Woo et al. in :cite:`woo2023etsformer`.
Parameters
----------
n_steps : int
The number of time steps in the time-series data sample.
n_features : int
The number of features in the time-series data sample.
anomaly_rate : float
The estimated anomaly rate in the dataset, within the range (0, 1). Used for thresholding.
n_encoder_layers : int
The number of layers in the ETSformer encoder.
n_decoder_layers : int
The number of layers in the ETSformer decoder.
d_model : int
The dimension of the model.
n_heads : int
The number of heads in each layer of ETSformer.
d_ffn : int
The dimension of the feed-forward network.
top_k : int
Top-K Fourier bases.
dropout : float, optional
The dropout rate for the model.
ORT_weight : float, optional
The weight for the ORT loss.
MIT_weight : float, optional
The weight for the MIT loss.
batch_size : int, optional
The number of samples per batch during training and evaluation.
epochs : int, optional
Total number of training epochs.
patience : int, optional
Number of epochs to wait for improvement before triggering early stopping.
training_loss : Criterion or type, optional
Loss function used during training. Defaults to MAE.
validation_metric : Criterion or type, optional
Metric used during validation. Defaults to MSE.
optimizer : Optimizer or type, optional
Optimizer used for training. Defaults to custom Adam optimizer.
num_workers : int, optional
Number of subprocesses used for data loading.
device : str, torch.device, or list, optional
Device(s) used for model training and inference.
saving_path : str, optional
Path to save model checkpoints and training logs.
model_saving_strategy : str or None, optional
Strategy to save models: one of {None, "best", "better", "all"}.
verbose : bool, optional
Whether to print training logs during execution.
"""
def __init__(
self,
n_steps: int,
n_features: int,
anomaly_rate: float,
n_encoder_layers: int,
n_decoder_layers: int,
d_model: int,
n_heads: int,
d_ffn: int,
top_k: int,
dropout: float = 0,
ORT_weight: float = 1,
MIT_weight: float = 1,
batch_size: int = 32,
epochs: int = 100,
patience: Optional[int] = None,
training_loss: Union[Criterion, type] = MAE,
validation_metric: Union[Criterion, type] = MSE,
optimizer: Union[Optimizer, type] = Adam,
num_workers: int = 0,
device: Optional[Union[str, torch.device, list]] = None,
saving_path: str = None,
model_saving_strategy: Optional[str] = "best",
verbose: bool = True,
):
super().__init__(
anomaly_rate=anomaly_rate,
training_loss=training_loss,
validation_metric=validation_metric,
batch_size=batch_size,
epochs=epochs,
patience=patience,
num_workers=num_workers,
device=device,
saving_path=saving_path,
model_saving_strategy=model_saving_strategy,
verbose=verbose,
)
# Store model hyperparameters
self.n_steps = n_steps
self.n_features = n_features
self.n_heads = n_heads
self.n_encoder_layers = n_encoder_layers
self.n_decoder_layers = n_decoder_layers
self.d_model = d_model
self.d_ffn = d_ffn
self.dropout = dropout
self.top_k = top_k
self.ORT_weight = ORT_weight
self.MIT_weight = MIT_weight
# Initialize the ETSformer model
self.model = _ETSformer(
n_steps=self.n_steps,
n_features=self.n_features,
n_encoder_layers=self.n_encoder_layers,
n_decoder_layers=self.n_decoder_layers,
d_model=self.d_model,
n_heads=self.n_heads,
d_ffn=self.d_ffn,
dropout=self.dropout,
top_k=self.top_k,
ORT_weight=self.ORT_weight,
MIT_weight=self.MIT_weight,
training_loss=self.training_loss,
validation_metric=self.validation_metric,
)
self._send_model_to_given_device()
self._print_model_size()
# Set up the optimizer
if isinstance(optimizer, Optimizer):
self.optimizer = optimizer
else:
self.optimizer = optimizer()
assert isinstance(self.optimizer, Optimizer)
self.optimizer.init_optimizer(self.model.parameters())
def _assemble_input_for_training(self, data: list) -> dict:
(
indices,
X,
missing_mask,
X_ori,
indicating_mask,
) = self._send_data_to_given_device(data)
return {
"X": X,
"missing_mask": missing_mask,
"X_ori": X_ori,
"indicating_mask": indicating_mask,
}
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, missing_mask = self._send_data_to_given_device(data)
return {
"X": X,
"missing_mask": missing_mask,
}
[docs]
def fit(
self,
train_set: Union[dict, str],
val_set: Optional[Union[dict, str]] = None,
file_type: str = "hdf5",
) -> None:
"""
Trains the ETSformer model on the given dataset.
"""
self.train_set = train_set
# Step 1: Wrap training data
train_dataset = DatasetForSAITS(train_set, return_X_ori=False, return_y=False, file_type=file_type)
train_dataloader = DataLoader(
train_dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=self.num_workers,
)
# Step 2: Wrap validation data if available
val_dataloader = 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_dataset = DatasetForSAITS(val_set, return_X_ori=True, return_y=False, file_type=file_type)
val_dataloader = DataLoader(
val_dataset,
batch_size=self.batch_size,
shuffle=False,
num_workers=self.num_workers,
)
# Step 3: Train the model
self._train_model(train_dataloader, val_dataloader)
# Step 4: Load best model
self.model.load_state_dict(self.best_model_dict)
# Step 5: Save the model if necessary
self._auto_save_model_if_necessary(confirm_saving=self.model_saving_strategy == "best")