openfl.pipelines.stc_pipeline.SparsityTransformer

class openfl.pipelines.stc_pipeline.SparsityTransformer(p=0.01)

Bases: Transformer

A transformer class to sparsify input data.

Class Attributes:

  • p (float) – The sparsity ratio.

  • lossy (bool) – A flag indicating if the transformation is lossy.

Methods

backward

Recover data array with the right shape and numerical type.

forward

Sparsify data and pass over only non-sparsified elements by reducing the array size.
backward(data, metadata, **kwargs)

Recover data array with the right shape and numerical type.

Parameters:

  • data – an numpy array with non-zero values.

  • metadata – dictionary to contain information for recovering back to original data array.

Returns:

recovered_data – an numpy array with original shape.

forward(data, **kwargs)

Sparsify data and pass over only non-sparsified elements by reducing the array size.

Parameters:

data – an numpy array from the model tensor_dict.

Returns:

  • sparse_data – a flattened, sparse representation of the input tensor.

  • metadata – dictionary to store a list of meta information.