def test(self):
"""
Test Data set
:param X:
:param y:
:return:
"""
return Validator.get_accuracy_with_confusion_matrix(self.model, self.X_test, self.Y_test)
def train(self):
iterations_count, iterations_per_epoch = Worker.get_iterations_count(self.X_train.shape[0], self.batch_size,
self.epochs_count)
for t in range(iterations_count):
self._send_pulse()
# Maybe print training loss
if self.debug and t % self.debug_every == 0:
print('loss: %f after (Iteration %d / %d)' % (
self.loss,t + 1, iterations_count,))
#Update learning rate after an epoch
epoch_end = (t + 1) % iterations_per_epoch == 0
if epoch_end:
self.epoch += 1
for k in range(len(self.model.layer_objs)):
self.W_configs[k].decay(self.lr_decay)
self.b_configs[k].decay(self.lr_decay)
#Print accuracy after each epoch
if epoch_end:
train_acc = Validator.get_accuracy(self.model, self.X_train, self.Y_train,
sample_size=1000)
val_acc = Validator.get_accuracy(self.model, self.X_val, self.Y_val)
if self.debug:
print('(Epoch %d / %d) train_acc: %f; val_acc: %f' % (
self.epoch, self.epochs_count, train_acc, val_acc))
if val_acc > self.best_val_acc:
self.best_val_acc = val_acc
self.best_Ws = []
self.best_bs = []
for layer_obj in self.model.layer_objs:
self.best_Ws.append(layer_obj.W)
self.best_bs.append(layer_obj.b)
for layer_id in range(len(self.model.layer_objs)):
self.model.layer_objs[layer_id].W = self.best_Ws[layer_id]
self.model.layer_objs[layer_id].b = self.best_bs[layer_id]