Class: DNN::Activations::SigmoidWithLoss

Inherits:

Layers::OutputLayer

• Object
• Layers::Layer
• Layers::OutputLayer
• DNN::Activations::SigmoidWithLoss

Defined in:

lib/dnn/core/activations.rb

Constant Summary

NMath =

Xumo::NMath

Instance Method Summary

• #backward(y) ⇒ Object
• #forward(x) ⇒ Object
• #initialize ⇒ SigmoidWithLoss constructor

  A new instance of SigmoidWithLoss.

• #loss(y) ⇒ Object

Methods inherited from Layers::Layer

#build, #built?, #prev_layer, #shape, #to_hash

Constructor Details

#initialize ⇒ SigmoidWithLoss

Returns a new instance of SigmoidWithLoss.

# File 'lib/dnn/core/activations.rb', line 244

def initialize
  @sigmoid = Sigmoid.new
end

Instance Method Details

#backward(y) ⇒ Object

# File 'lib/dnn/core/activations.rb', line 252

def backward(y)
  @out - y
end

#forward(x) ⇒ Object

# File 'lib/dnn/core/activations.rb', line 248

def forward(x)
  @out = @sigmoid.forward(x)
end

#loss(y) ⇒ Object

# File 'lib/dnn/core/activations.rb', line 256

def loss(y)
  batch_size = y.shape[0]
  -(y * NMath.log(@out + 1e-7) + (1 - y) * NMath.log(1 - @out + 1e-7)).sum / batch_size + lasso + ridge
end