def main():
# --- 引数処理 ---
args = ArgParser()
print(args.data_type)
print(args.dataset_dir)
print(args.model_type)
if (args.data_type == "MNIST"):
dataset = DataLoaderMNIST(args.dataset_dir)
print(dataset.train_images.shape)
print(dataset.train_labels.shape)
print(dataset.test_images.shape)
print(dataset.test_labels.shape)
x_train = dataset.train_images / 255
y_train = dataset.train_labels
x_test = dataset.test_images / 255
y_test = dataset.test_labels
output_dims = dataset.output_dims
elif (args.data_type == "CIFAR-10"):
dataset = DataLoaderCIFAR10(args.dataset_dir)
print(dataset.train_images.shape)
print(dataset.train_labels.shape)
print(dataset.test_images.shape)
print(dataset.test_labels.shape)
x_train = dataset.train_images / 255
y_train = dataset.train_labels
x_test = dataset.test_images / 255
y_test = dataset.test_labels
output_dims = dataset.output_dims
else:
print('[ERROR] Unknown data_type: {}'.format(args.data_type))
quit()
if (args.model_type == 'MLP'):
trainer = TrainerMLP(dataset.train_images.shape[1:], output_dir='./output')
trainer.fit(x_train, y_train, x_test=x_test, y_test=y_test)
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
elif (args.model_type == 'CNN'):
trainer = TrainerCNN(dataset.train_images.shape[1:], output_dir='./output')
trainer.fit(x_train, y_train, x_test=x_test, y_test=y_test)
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
elif (args.model_type == 'ResNet'):
trainer = TrainerResNet(dataset.train_images.shape[1:], output_dims, output_dir='./output')
trainer.fit(x_train, y_train, x_test=x_test, y_test=y_test)
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
else:
print('[ERROR] Unknown model_type: {}'.format(args.model_type))
quit()
return
def main():
# --- NumPy配列形状表示 ---
def print_ndarray_shape(ndarr):
if (ndarr is not None):
print(ndarr.shape)
else:
pass
return
# --- 引数処理 ---
args = ArgParser()
print('[INFO] Arguments')
print(' * args.data_type = {}'.format(args.data_type))
print(' * args.dataset_dir = {}'.format(args.dataset_dir))
print(' * args.model_type = {}'.format(args.model_type))
print(' * args.data_augmentation = {}'.format(args.data_augmentation))
print(' * args.optimizer = {}'.format(args.optimizer))
print(' * args.batch_size = {}'.format(args.batch_size))
print(' * args.initializer = {}'.format(args.initializer))
print(' * args.result_dir = {}'.format(args.result_dir))
# --- Data Augmentationパラメータを辞書型に変換 ---
if (args.data_augmentation is not None):
dict_keys = [
'rotation_range', 'width_shift_range', 'height_shift_range',
'horizontal_flip'
]
df_da_params = pd.read_csv(io.StringIO(args.data_augmentation),
header=None,
skipinitialspace=True).values[0]
data_augmentation = {}
for (key, da_param) in zip(dict_keys, df_da_params):
data_augmentation[key] = da_param
else:
data_augmentaion = None
if (args.data_type == "MNIST"):
dataset = DataLoaderMNIST(args.dataset_dir, validation_split=0.2)
print_ndarray_shape(dataset.train_images)
print_ndarray_shape(dataset.train_labels)
print_ndarray_shape(dataset.validation_images)
print_ndarray_shape(dataset.validation_labels)
print_ndarray_shape(dataset.test_images)
print_ndarray_shape(dataset.test_labels)
if (dataset.train_images is not None):
x_train = dataset.train_images / 255
else:
x_train = None
y_train = dataset.train_labels
if (dataset.validation_images is not None):
x_val = dataset.validation_images / 255
else:
x_val = None
y_val = dataset.validation_labels
if (dataset.test_images is not None):
x_test = dataset.test_images / 255
else:
x_test = None
y_test = dataset.test_labels
output_dims = dataset.output_dims
elif (args.data_type == "CIFAR-10"):
dataset = DataLoaderCIFAR10(args.dataset_dir, validation_split=0.2)
print_ndarray_shape(dataset.train_images)
print_ndarray_shape(dataset.train_labels)
print_ndarray_shape(dataset.validation_images)
print_ndarray_shape(dataset.validation_labels)
print_ndarray_shape(dataset.test_images)
print_ndarray_shape(dataset.test_labels)
if (dataset.train_images is not None):
x_train = dataset.train_images / 255
else:
x_train = None
y_train = dataset.train_labels
if (dataset.validation_images is not None):
x_val = dataset.validation_images / 255
else:
x_val = None
y_val = dataset.validation_labels
if (dataset.test_images is not None):
x_test = dataset.test_images / 255
else:
x_test = None
y_test = dataset.test_labels
output_dims = dataset.output_dims
else:
print('[ERROR] Unknown data_type: {}'.format(args.data_type))
quit()
if (args.model_type == 'MLP'):
trainer = TrainerMLP(dataset.train_images.shape[1:],
output_dir=args.result_dir,
optimizer=args.optimizer,
initializer=args.initializer)
trainer.fit(x_train,
y_train,
x_val=x_val,
y_val=y_val,
x_test=x_test,
y_test=y_test,
batch_size=args.batch_size,
da_params=data_augmentation)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
elif (args.model_type == 'SimpleCNN'):
trainer = TrainerCNN(dataset.train_images.shape[1:],
output_dir=args.result_dir,
optimizer=args.optimizer,
initializer=args.initializer)
trainer.fit(x_train,
y_train,
x_val=x_val,
y_val=y_val,
x_test=x_test,
y_test=y_test,
batch_size=args.batch_size,
da_params=data_augmentation)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
elif (args.model_type == 'SimpleResNet'):
trainer = TrainerResNet(dataset.train_images.shape[1:],
output_dims,
output_dir=args.result_dir,
optimizer=args.optimizer,
initializer=args.initializer)
trainer.fit(x_train,
y_train,
x_val=x_val,
y_val=y_val,
x_test=x_test,
y_test=y_test,
batch_size=args.batch_size,
da_params=data_augmentation)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
else:
print('[ERROR] Unknown model_type: {}'.format(args.model_type))
quit()
return
def main():
# --- NumPy配列形状表示 ---
def print_ndarray_shape(ndarr):
if (ndarr is not None):
print(ndarr.shape)
else:
pass
return
# --- 引数処理 ---
args = ArgParser()
print('[INFO] Arguments')
print(' * args.data_type = {}'.format(args.data_type))
print(' * args.dataset_dir = {}'.format(args.dataset_dir))
print(' * args.model_type = {}'.format(args.model_type))
print(' * args.data_augmentation = {}'.format(args.data_augmentation))
print(' * args.optimizer = {}'.format(args.optimizer))
print(' * args.result_dir = {}'.format(args.result_dir))
if (args.data_type == "MNIST"):
dataset = DataLoaderMNIST(args.dataset_dir, validation_split=0.2)
print_ndarray_shape(dataset.train_images)
print_ndarray_shape(dataset.train_labels)
print_ndarray_shape(dataset.validation_images)
print_ndarray_shape(dataset.validation_labels)
print_ndarray_shape(dataset.test_images)
print_ndarray_shape(dataset.test_labels)
if (dataset.train_images is not None):
x_train = dataset.train_images / 255
else:
x_train = None
y_train = dataset.train_labels
if (dataset.validation_images is not None):
x_val = dataset.validation_images / 255
else:
x_val = None
y_val = dataset.validation_labels
if (dataset.test_images is not None):
x_test = dataset.test_images / 255
else:
x_test = None
y_test = dataset.test_labels
output_dims = dataset.output_dims
elif (args.data_type == "CIFAR-10"):
dataset = DataLoaderCIFAR10(args.dataset_dir, validation_split=0.2)
print_ndarray_shape(dataset.train_images)
print_ndarray_shape(dataset.train_labels)
print_ndarray_shape(dataset.validation_images)
print_ndarray_shape(dataset.validation_labels)
print_ndarray_shape(dataset.test_images)
print_ndarray_shape(dataset.test_labels)
if (dataset.train_images is not None):
x_train = dataset.train_images / 255
else:
x_train = None
y_train = dataset.train_labels
if (dataset.validation_images is not None):
x_val = dataset.validation_images / 255
else:
x_val = None
y_val = dataset.validation_labels
if (dataset.test_images is not None):
x_test = dataset.test_images / 255
else:
x_test = None
y_test = dataset.test_labels
output_dims = dataset.output_dims
else:
print('[ERROR] Unknown data_type: {}'.format(args.data_type))
quit()
if (args.model_type == 'MLP'):
trainer = TrainerMLP(dataset.train_images.shape[1:],
output_dir=args.result_dir,
optimizer=args.optimizer)
trainer.fit(x_train,
y_train,
x_val=x_val,
y_val=y_val,
x_test=x_test,
y_test=y_test,
da_enable=args.data_augmentation)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
elif (args.model_type == 'SimpleCNN'):
trainer = TrainerCNN(dataset.train_images.shape[1:],
output_dir=args.result_dir,
optimizer=args.optimizer)
trainer.fit(x_train,
y_train,
x_val=x_val,
y_val=y_val,
x_test=x_test,
y_test=y_test,
da_enable=args.data_augmentation)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
elif (args.model_type == 'SimpleResNet'):
trainer = TrainerResNet(dataset.train_images.shape[1:],
output_dims,
output_dir=args.result_dir,
optimizer=args.optimizer)
trainer.fit(x_train,
y_train,
x_val=x_val,
y_val=y_val,
x_test=x_test,
y_test=y_test,
da_enable=args.data_augmentation)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
else:
print('[ERROR] Unknown model_type: {}'.format(args.model_type))
quit()
return
def main():
# --- NumPy配列形状表示 ---
def print_ndarray_shape(ndarr):
if (ndarr is not None):
print(ndarr.shape)
else:
pass
return
# --- 引数処理 ---
args = ArgParser()
print('[INFO] Arguments')
print(' * args.fifo = {}'.format(args.fifo))
print(' * args.data_type = {}'.format(args.data_type))
print(' * args.dataset_dir = {}'.format(args.dataset_dir))
print(' * args.model_type = {}'.format(args.model_type))
print(' * args.data_augmentation = {}'.format(args.data_augmentation))
print(' * args.optimizer = {}'.format(args.optimizer))
print(' * args.batch_size = {}'.format(args.batch_size))
print(' * args.initializer = {}'.format(args.initializer))
print(' * args.data_norm = {}'.format(args.data_norm))
print(' * args.dropout_rate = {}'.format(args.dropout_rate))
print(' * args.loss_func = {}'.format(args.loss_func))
print(' * args.epochs = {}'.format(args.epochs))
print(' * args.result_dir = {}'.format(args.result_dir))
# --- Data Augmentationパラメータを辞書型に変換 ---
if (args.data_augmentation is not None):
dict_keys = ['rotation_range', 'width_shift_range', 'height_shift_range', 'zoom_range', 'channel_shift_range', 'horizontal_flip']
df_da_params = pd.read_csv(io.StringIO(args.data_augmentation), header=None, skipinitialspace=True).values[0]
data_augmentation = {}
for (key, da_param) in zip(dict_keys, df_da_params):
data_augmentation[key] = da_param
else:
data_augmentaion = None
if (args.loss_func == "sparse_categorical_crossentropy"):
one_hot = False
else:
one_hot = True
if (args.data_type == "MNIST"):
dataset = DataLoaderMNIST(args.dataset_dir, validation_split=0.2, one_hot=one_hot)
elif (args.data_type == "CIFAR-10"):
dataset = DataLoaderCIFAR10(args.dataset_dir, validation_split=0.2, one_hot=one_hot)
else:
print('[ERROR] Unknown data_type: {}'.format(args.data_type))
quit()
print_ndarray_shape(dataset.train_images)
print_ndarray_shape(dataset.train_labels)
print_ndarray_shape(dataset.validation_images)
print_ndarray_shape(dataset.validation_labels)
print_ndarray_shape(dataset.test_images)
print_ndarray_shape(dataset.test_labels)
x_train, x_val, x_test = dataset.normalization(args.data_norm)
y_train = dataset.train_labels
y_val = dataset.validation_labels
y_test = dataset.test_labels
output_dims = dataset.output_dims
if (args.model_type == 'MLP'):
trainer = TrainerMLP(dataset.train_images.shape[1:], output_dir=args.result_dir,
optimizer=args.optimizer, initializer=args.initializer)
elif (args.model_type == 'SimpleCNN'):
trainer = TrainerCNN(dataset.train_images.shape[1:], output_dir=args.result_dir,
optimizer=args.optimizer, loss=args.loss_func, initializer=args.initializer)
elif (args.model_type == 'DeepCNN'):
trainer = TrainerCNN(dataset.train_images.shape[1:], output_dir=args.result_dir,
optimizer=args.optimizer, loss=args.loss_func, initializer=args.initializer, model_type='deep_model')
elif (args.model_type == 'SimpleResNet'):
trainer = TrainerResNet(dataset.train_images.shape[1:], output_dims, output_dir=args.result_dir,
model_type='custom',
optimizer=args.optimizer, loss=args.loss_func, initializer=args.initializer, dropout_rate=args.dropout_rate)
elif (args.model_type == 'DeepResNet'):
trainer = TrainerResNet(dataset.train_images.shape[1:], output_dims, output_dir=args.result_dir,
model_type='custom_deep',
optimizer=args.optimizer, loss=args.loss_func, initializer=args.initializer, dropout_rate=args.dropout_rate)
else:
print('[ERROR] Unknown model_type: {}'.format(args.model_type))
quit()
trainer.fit(args.fifo, x_train, y_train, x_val=x_val, y_val=y_val, x_test=x_test, y_test=y_test,
batch_size=args.batch_size, da_params=data_augmentation, epochs=args.epochs)
trainer.save_model()
predictions = trainer.predict(x_test)
print('\nPredictions(shape): {}'.format(predictions.shape))
return