def supervised_cost_train(self, targets):
# todo may have to do something more around the labelled vs unlabelled data
labeled_activations_corrupted = labeled(
self.input_layer.activation_train
) #tf.slice(self.activation, [0, 0], tf.shape(targets))
return -tf.reduce_mean(
tf.reduce_sum(targets * tf.log(labeled_activations_corrupted), 1))
def supervised_cost_train(self, targets):
# todo may have to do something more around the labelled vs unlabelled data
labeled_activations_corrupted = labeled(self.input_layer.activation_train) #tf.slice(self.activation, [0, 0], tf.shape(targets))
return -tf.reduce_mean(tf.reduce_sum(targets * tf.log(labeled_activations_corrupted), 1))
def activation_train(self):
return labeled(self.input_layer.activation_predict)
def activation_train(self):
return labeled(self.input_layer.activation_predict)
