phitodeep.layers.activation

Classes

ReLu	Base class for all layers in the network.
LeakyReLu	Base class for all layers in the network.
GELU	Base class for all layers in the network.
Swish	Base class for all layers in the network.
Sigmoid	Base class for all layers in the network.
Tanh	Base class for all layers in the network.
Softmax	Base class for all layers in the network.
ELU	Base class for all layers in the network.

Module Contents

class phitodeep.layers.activation.ReLu

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

forward(X)

backward(dL_dZ)

Backpropagate through ReLU activation. ReLU derivative: 1 if X > 0, else 0

copy()

class phitodeep.layers.activation.LeakyReLu(alpha=0.01)

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

alpha = 0.01

forward(X)

backward(dL_dZ)

Backward pass through the block.

Parameters:

dL_dZ – gradient of loss w.r.t. output of this block

Returns:

gradient of loss w.r.t. input (to pass to previous layer)

Return type:

dL_dX

copy()

class phitodeep.layers.activation.GELU

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

constant = 0.044715

forward(X)

backward(dL_dZ)

Backward pass through the block.

Parameters:

dL_dZ – gradient of loss w.r.t. output of this block

Returns:

gradient of loss w.r.t. input (to pass to previous layer)

Return type:

dL_dX

copy()

class phitodeep.layers.activation.Swish

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

forward(X)

backward(dL_dZ)

Backward pass through the block.

Parameters:

dL_dZ – gradient of loss w.r.t. output of this block

Returns:

gradient of loss w.r.t. input (to pass to previous layer)

Return type:

dL_dX

copy()

class phitodeep.layers.activation.Sigmoid

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

forward(X)

backward(dL_dZ)

Backpropagate through Sigmoid activation. Sigmoid derivative: sigmoid(Z) * (1 - sigmoid(Z))

copy()

class phitodeep.layers.activation.Tanh

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

forward(X)

backward(dL_dZ)

Backpropagate through Tanh activation. Tanh derivative: 1 - tanh(Z)^2

copy()

class phitodeep.layers.activation.Softmax

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

forward(X)

backward(dL_dZ)

Backpropagate through Softmax activation. When paired with CategoricalCrossEntropy, the combined gradient (y_pred - one_hot(y_true)) / N is computed entirely in the loss, so this layer is a straight pass-through.

copy()

class phitodeep.layers.activation.ELU(alpha=1.0)

Bases: phitodeep.layers.base.Layer

Base class for all layers in the network.

alpha_activation = 1.0

forward(X)

backward(dL_dZ)

Backpropagate through ELU activation. ELU derivative: 1 if X > 0, else alpha * exp(X)

copy()