reducers

class torchdrift.reducers.PCAReducer(n_components: int = 2)

Reduce dimensions using PCA.

This nn.Modue subclass reduces the dimensions of the inputs specified by n_components.

The input is a 2-dimensional Tensor of size batch x features, the output is a Tensor of size batch x n_components.

fit(x: torch.Tensor) → torch.Tensor
class torchdrift.reducers.Reducer

Base class for reducers.

This is a torch.nn.Module with an additional fit method. The usual forward is for testing after fitting.

fit(x: torch.Tensor) → torch.Tensor

Fits the reducer to reference data x and returns the reduced data.

Override this in your reducer implementation.

forward(x: torch.Tensor) → torch.Tensor

Reduces the input x (in testing) and returns the reduced data.

Override this in your reducer implementation.