amatorch.normalization

Functions to normalize tensors in different ways.

Functions

unit_norm(stimuli[, c50])

Normalize stimuli to have a norm less than or equal to 1.

unit_norm_channels(stimuli[, c50])

Normalize stimuli to have a norm less than or equal to 1, normalizing each channel separately.
amatorch.normalization.unit_norm(stimuli, c50=0)

Normalize stimuli to have a norm less than or equal to 1.

Channels are normalized by their aggregated offsetted norm (i.e., || sum of squares + c50 ||).

Parameters:

  • stimuli (torch.Tensor) – Stimuli tensor of shape (n_stim, n_channels, n_dim).

  • c50 (torch.Tensor, optional) – Offset constant added to the sum of squares, by default torch.as_tensor(0).

Returns:

Normalized stimuli tensor of shape (n_stim, n_channels, n_dim).

Return type:

torch.Tensor

amatorch.normalization.unit_norm_channels(stimuli, c50=0)

Normalize stimuli to have a norm less than or equal to 1, normalizing each channel separately.

Each channel is normalized separately by its offsetted norm (i.e., sqrt( sum of squares + c50)), and both channels are normalized by the square root of the number of channels.

Parameters:

  • stimuli (torch.Tensor) – Stimuli tensor of shape (n_stim, n_channels, n_dim).

  • c50 (torch.Tensor, optional) – Offset constant added to the sum of squares, by default torch.as_tensor(0).

Returns:

Normalized stimuli tensor of shape (n_stim, n_channels, n_dim).

Return type:

torch.Tensor