lwthatcher / mlp Goto Github PK

The ReLU activation function cannot output a negative value. Thus when using the ReLU you can never have negative output for your network.

Right now the predict method seems to be doing regression, but not classification.
There should be an option to predict classifications.
.

Implement the basic sklearn interfaces for the NN:

Implement:

• BaseEstimator
• ClassifierMixin
• RegressorMixin?

Right now our activation functions' f_prime() function modify the array in-place, however it probably should return a new array and not modify the one given, since we will still need the output values for the weight update.

Progress

• modify ReLU f-prime function
• modify Sigmoid f-prime function
• add unit tests