class main():
def __init__(self):
self.train_img, self.train_lbl = load_data(1)
self.input_shape = (28, 28, 1)
# Hyper parameter values obtained after tuning using gird search
self.eta = 1e-3
self.epochs = 20
self.batch_size = 64
self.kernel_size = (4, 4)
self.model = CNN()
def train(self, activation):
history = {}
if activation is None:
activation = []
activation.append('relu')
for act in activation:
print("\nTraining for {} activation".format(act))
self.model.create_classifier(input_shape=self.input_shape,
activation=act,
eta=self.eta,
kernel_size=self.kernel_size)
history[act] = self.model.train_classifier(
self.train_img,
self.train_lbl,
epochs=self.epochs,
batch_size=self.batch_size,
activation=act)
#summarize history for accuracy
plt.figure(str(act) + ' acc')
plt.plot(history[act].history['acc'])
plt.plot(history[act].history['val_acc'])
plt.title(str(act) + ' model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'val'], loc='lower right')
plt.savefig(str(act) + '_accuracy.png')
# summarize history for loss
plt.figure(str(act) + ' loss')
plt.plot(history[act].history['loss'])
plt.plot(history[act].history['val_loss'])
plt.title(str(act) + ' model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'val'], loc='upper right')
plt.savefig(str(act) + '_loss.png')
legend = []
for values in history:
plt.figure('accuracy')
plt.plot(history[values].history['acc'])
legend.append(values)
plt.xlabel('epoch')
plt.ylabel('accuracy')
plt.legend(legend, loc='lower right')
plt.title('Accuracies for different activations')
plt.savefig('combined train accuracy.png')
for values in history:
plt.figure('val accuracy')
plt.plot(history[values].history['val_acc'])
plt.xlabel('epoch')
plt.ylabel('accuracy')
plt.legend(legend, loc='lower right')
plt.title('Validation Accuracies for different activations')
plt.savefig('combined val accuracy.png')
for values in history:
plt.figure('loss')
plt.plot(history[values].history['loss'])
plt.xlabel('epoch')
plt.ylabel('loss')
plt.legend(legend, loc='upper right')
plt.title('Loss for different activations')
plt.savefig('combined train loss.png')
for values in history:
plt.figure('val loss')
plt.plot(history[values].history['val_loss'])
plt.xlabel('epoch')
plt.ylabel('loss')
plt.legend(legend, loc='upper right')
plt.title('Validation loss for different activations')
plt.savefig('combined val loss.png')
def tune(self):
self.model.create_classifier(input_shape=self.input_shape,
eta=self.eta,
kernel_size=self.kernel_size)
grid_result = self.model.tune_hyperparameters(self.train_img,
self.train_lbl)
def test(self, activation):
if activation is None:
activation = []
activation.append('relu')
test_img, test_lbl = load_data(0)
for act in activation:
print("Test for " + act + ":")
test_accuracy, test_loss, predictions = self.model.test_classifier(
test_img, test_lbl, act)
print("\n")
predictions = np.round(predictions)
predictions = predictions.astype(int)
df = pd.DataFrame(predictions)
df.to_csv("mnist.csv", header=None, index=None)
