def errors(self, output, labels):
""" Calcultes the errors for the output and hidden layers"""
labels = Matrix.from_array(labels)
output_errors = labels.add(output.negative)
hidden_errors = Matrix.matmul(self.ho_weights.T, output_errors)
return hidden_errors, output_errors
def train(self, inputs, labels, stocastic=True):
""" Trains the neural net, with the option of stocastic
or batch training
"""
# Converting inputs to matrix object
inputs = Matrix.from_array(inputs)
hidden, output = self.feed_forward(inputs)
if stocastic:
hidden_gradient, delta_w_ih, output_gradient, delta_w_ho = \
self.back_propogate(inputs, hidden, output, labels)
self.output_bias.add(output_gradient, inplace=True)
self.ho_weights.add(delta_w_ho, inplace=True)
self.hidden_bias.add(hidden_gradient, inplace=True)
self.ih_weights.add(delta_w_ih, inplace=True)
else:
return self.back_propogate(inputs, hidden, output, labels)
def predict(self, inputs):
""" Method to test the neural net for a given input"""
inputs = Matrix.from_array(inputs)
return self.feed_forward(inputs)[1].data[0][0]