espnet2.asr.state_spaces.components.TSNormalization

source

class espnet2.asr.state_spaces.components.TSNormalization(method, horizon)

Bases: Module

Time Series Normalization module.

This module normalizes time series data using specified methods over a defined horizon. It supports normalization by either the mean or the last value within the specified horizon. The input tensor is expected to have the shape (B, L, D), where B is the batch size, L is the sequence length, and D is the number of features.

method

The normalization method to apply. Can be “mean” or “last”.

• Type: str

horizon

The number of time steps to consider for normalization.

• Type: int

• Parameters:

  • method (str) – The normalization method, either “mean” or “last”.
  • horizon (int) – The number of time steps to use for normalization.

• Returns: The normalized tensor with the same shape as the input.

• Return type: Tensor

####### Examples

>>> normalization = TSNormalization(method="mean", horizon=3)
>>> input_tensor = torch.tensor([[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]])
>>> normalized_tensor = normalization(input_tensor)
>>> print(normalized_tensor)
tensor([[[0.3333, 0.6667],
          [1.0000, 1.3333],
          [1.6667, 2.0000]]])

>>> normalization_last = TSNormalization(method="last", horizon=1)
>>> normalized_tensor_last = normalization_last(input_tensor)
>>> print(normalized_tensor_last)
tensor([[[1.0, 2.0],
          [1.5, 2.0],
          [2.5, 3.0]]])

NOTE

If the input tensor does not have the expected shape, the behavior of this module may be undefined.

• Raises:ValueError – If the input tensor does not have at least 2 dimensions.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x)

Time Series Normalization module.

This module normalizes time series data based on the specified method and horizon. The normalization is performed on the input tensor, which is expected to have a shape of (B, L, D), where B is the batch size, L is the length of the sequence, and D is the number of features.

method

The normalization method to use. Options are “mean” or “last”.

• Type: str

horizon

The number of timesteps to consider for normalization.

• Type: int

• Parameters:

  • method (str) – The normalization method to use (“mean” or “last”).
  • horizon (int) – The number of timesteps to consider for normalization.

• Returns: The normalized tensor.

• Return type: Tensor

####### Examples

>>> normalization = TSNormalization(method="mean", horizon=5)
>>> x = torch.randn(10, 20, 3)  # A batch of 10 sequences, each of length 20 with 3 features
>>> normalized_x = normalization(x)
>>> print(normalized_x.shape)  # Output: torch.Size([10, 20, 3])

>>> normalization_last = TSNormalization(method="last", horizon=5)
>>> normalized_last_x = normalization_last(x)
>>> print(normalized_last_x.shape)  # Output: torch.Size([10, 20, 3])

NOTE

Ensure that the input tensor has the correct shape (B, L, D) before using this module.