def demo_cnn_aug():
X, y, w = load2d()
y = flatten_except_1dim(y,ndim=3)
w = flatten_except_1dim(w,ndim=2)
X_train, X_val, y_train, y_val, w_train, w_val = train_test_split(X, y, w, test_size=0.2, random_state=42)
model = CNN()
flip_indices = [
(0, 2), (1, 3),
(4, 8), (5, 9), (6, 10), (7, 11),
(12, 16), (13, 17), (14, 18), (15, 19),
(22, 24), (23, 25),
]
hist = model.fit_generator(FlippedImageDataGenerator(X_train, y_train, w_train, 32, flip_indices=flip_indices),
# steps_per_epoch=X_train.shape[0],
epochs=EPOCH,
validation_data=(X_val, y_val, w_val)
)
model.save('./models/cnn_aug_weighted_model.h5')
X_test,_,_ = load2d(test=True)
df_y_pred = predict_single(model, X_test)
prepare_submission(df_y_pred,"cnn_aug_weighted")
def train_CNN(data_portion=X.shape[0], epochs=20, ensemble=False):
model = CNN(im_shape)
# epochs = 50
batch = 100
if augment:
history = model.fit_generator(datagen.flow(train_X, train_y),
epochs=epochs,
steps_per_epoch=data_portion // 32,
verbose=2,
validation_data=[val_X, val_y],
callbacks=[annealer])
else:
history = model.fit(train_X[:data_portion],
train_y[:data_portion],
batch,
epochs,
verbose=2,
validation_split=0.3,
callbacks=[annealer])
# print(model.summary())
if not ensemble:
# print('saving')
# model.save('saved_models/my_model.h5')
# print('done')
fig = plt.figure(figsize=(15, 5))
plt.plot(history.history['val_acc'])
plt.xlabel('epoch')
plt.ylabel('val_accuracy')
plt.xlim(0, epochs)
plt.grid()
plt.ylim(.97, 1)
plt.title('Max val_acc: ' + str(max(history.history['val_acc'])))
fig.savefig('plots/testing.png')
return model
checkpoint = ModelCheckpoint(paramfile,
monitor='val_acc',
save_best_only=True,
save_weights_only=True,
verbose=0,
mode='max')
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
print('Start training...')
if args.model != 'DNN':
model.fit_generator(datagen.flow(train_x,
train_y,
batch_size=args.batch),
steps_per_epoch=5 * num_batches,
epochs=args.epoch,
verbose=1,
validation_data=(valid_x, valid_y),
workers=8,
callbacks=[checkpoint, earlystopping, csvlogger])
else:
model.fit(train_x,
train_y,
batch_size=args.batch,
epochs=args.epoch,
verbose=1,
validation_data=(valid_x, valid_y),
shuffle=True,
callbacks=[checkpoint, earlystopping, csvlogger])
if args.test == True:
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max')
lr_reducer = ReduceLROnPlateau(monitor='val_loss',
factor=0.35,
patience=3,
verbose=1)
# lr_finder = LRFinder(min_lr=1e-6, max_lr=1e-2, steps_per_epoch=data_portion//batch_size, epochs=epochs)
model = CNN(input_shape)
history = model.fit_generator(
train_generator,
steps_per_epoch=int(data_portion * 0.8) // batch_size,
epochs=epochs,
validation_data=val_generator,
validation_steps=int(data_portion * 0.2) // batch_size,
callbacks=[lr_reducer, checkpoint],
verbose=1)
print(model.summary())
model.save('my_model2.h5')
# lr_finder.plot_loss()
plt.subplot(1, 2, 1)
plt.plot(history.history['val_acc'])
plt.xlim(0, epochs)
plt.ylim(0.5, 1)
plt.xlabel('Epoch')
plt.ylabel('Validation Accuracy')
plt.title(max(history.history['val_acc']))
train_generator = ImageDataGenerator(rotation_range=10,
width_shift_range=0.05,
height_shift_range=0.05,
horizontal_flip=True,
shear_range=0.2,
zoom_range=0.2).flow(
x_train,
y_train,
batch_size=opt.batch_size)
valid_generator = ImageDataGenerator().flow(x_valid,
y_valid,
batch_size=opt.batch_size)
history_fer2013 = model.fit_generator(
train_generator,
steps_per_epoch=len(y_train) // opt.batch_size,
epochs=opt.epochs,
validation_data=valid_generator,
validation_steps=len(y_valid) // opt.batch_size,
callbacks=callback)
his = history_fer2013
# test
pred = model.predict(x_test)
pred = np.argmax(pred, axis=1)
print("test accuacy",
np.sum(pred.reshape(-1) == y_test.reshape(-1)) / y_test.shape[0])
elif opt.dataset == "jaffe":
expressions, x, y = Jaffe().gen_train()
y = to_categorical(y).reshape(y.shape[0], -1)
# 为了统一几个数据集，必须增加一列为0的
