Calculate the delta of this iteration.
Calculate the delta of this iteration. The input of the layer in forward phase can be restored from ActivationStack. It returns the delta of input layer of this layer and the delta of coefficient and intercept parameter.
Returns the form for creating tmp model used while pretraining The layer used as prototype for creating tmp model.
Returns the form for creating tmp model used while pretraining The layer used as prototype for creating tmp model. Only necessary fields are input size, output size and id.
A new pretrain layer that is reversed output and input. It is used mainly for keeping bias value while pretraining.
Decode hidden layer value to visible layer
Decode hidden layer value to visible layer
Encode the input to hidden layer
Encode the input to hidden layer
Calculate the error of output layer between label data and prediction.
Calculate the error of output layer between label data and prediction.
Calculate the output corresponding given input.
Calculate the output corresponding given input. Input is given as a top of ActivationStack.
The output tuple of the layer. First value of the tuple represents the raw output, the second is applied activation function of the layer.
Pretraining with input data to the layer.
Pretraining with input data to the layer. We can assume AutoEncoder can inherit this class. It returns the gradient of weight and bias of the layer.
(dWeight1, dBias1): Hidden Layer Gradient (dWeight2, dBias2): Visible Layer Gradient (Hidden Layer Gradient, Visible Layer Gradient)