merging

class sparx.merging.GlobalMerger

Bases: Merger

Merges the given model using the global merge technique.

static merge(mlp: FFNN, cluster_labels: ndarray) → FFNN

Applies merge algorithm to compress layer into clustered-layer

Parameters:

• mlp (FFNN) – the MLP model to merge on

• cluster_labels (np.ndarray) – cluster labels for each layer of the network

Raises:

NotImplementedError –

Returns:

resulting MLP model with clustered layers

Return type:

FFNN

class sparx.merging.LocalMerger

Bases: Merger

Merges the given model using the local merge technique.

static merge(mlp: FFNN, cluster_labels: ndarray) → FFNN

Applies merge algorithm to compress layer into clustered-layer

Parameters:

• mlp (FFNN) – the MLP model to merge on

• cluster_labels (np.ndarray) – cluster labels for each layer of the network

Raises:

NotImplementedError –

Returns:

resulting MLP model with clustered layers

Return type:

FFNN

class sparx.merging.Merger

Bases: object

Abstract class for merging algorithms.

abstract static merge(mlp: FFNN, cluster_labels: ndarray) → FFNN

Applies merge algorithm to compress layer into clustered-layer

Parameters:

• mlp (FFNN) – the MLP model to merge on

• cluster_labels (np.ndarray) – cluster labels for each layer of the network

Raises:

NotImplementedError –

Returns:

resulting MLP model with clustered layers

Return type:

FFNN