p2pfl.learning.compression.base_compression_strategy module

Compression strategy interface.

class p2pfl.learning.compression.base_compression_strategy.ByteCompressor[source]

Bases: CompressionStrategy

Subclass for byte compression strategies.

abstract apply_strategy(data)[source]

Apply strategy to the parameters.

Parameters:

data (bytes)

Return type:

bytes

abstract reverse_strategy(data)[source]

Reverse the strategy.

Parameters:

data (bytes)

Return type:

bytes

class p2pfl.learning.compression.base_compression_strategy.CompressionStrategy[source]

Bases: ABC

Abstract class for optimization strategies.

abstract apply_strategy(*args, **kwargs)[source]

Apply strategy to the parameters.

Return type:

Any

abstract reverse_strategy(*args, **kwargs)[source]

Reverse the strategy.

Return type:

Any

class p2pfl.learning.compression.base_compression_strategy.TensorCompressor[source]

Bases: CompressionStrategy

Subclass for tensor compression strategies.

abstract apply_strategy(params)[source]

Apply strategy to the parameters.

Parameters:

params (list[ndarray])

Return type:

tuple[list[ndarray], dict]

abstract reverse_strategy(params, additional_info)[source]

Reverse the strategy.

Parameters:

  • params (list[ndarray])

  • additional_info (dict)

Return type:

list[ndarray]