_archiving/contribution/daekeun-ml/tensorflow-in-sagemaker-workshop/training_script/cifar10_keras_sm_solution.py [31:208]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - sess = tf.Session() K.set_session(sess) logging.getLogger().setLevel(logging.INFO) tf.logging.set_verbosity(tf.logging.INFO) HEIGHT = 32 WIDTH = 32 DEPTH = 3 NUM_CLASSES = 10 NUM_DATA_BATCHES = 5 NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 10000 * NUM_DATA_BATCHES INPUT_TENSOR_NAME = 'inputs_input' # needs to match the name of the first layer + "_input" def keras_model_fn(learning_rate, weight_decay, optimizer, momentum): """keras_model_fn receives hyperparameters from the training job and returns a compiled keras model. The model will be transformed into a TensorFlow Estimator before training and it will be saved in a TensorFlow Serving SavedModel at the end of training. Args: hyperparameters: The hyperparameters passed to the SageMaker TrainingJob that runs your TensorFlow training script. Returns: A compiled Keras model """ model = Sequential() model.add(Conv2D(32, (3, 3), padding='same', name='inputs', input_shape=(HEIGHT, WIDTH, DEPTH))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(Conv2D(32, (3, 3))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.2)) model.add(Conv2D(64, (3, 3), padding='same')) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(Conv2D(64, (3, 3))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.3)) model.add(Conv2D(128, (3, 3), padding='same')) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(Conv2D(128, (3, 3))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.4)) model.add(Flatten()) model.add(Dense(512)) model.add(Activation('relu')) model.add(Dropout(0.5)) model.add(Dense(NUM_CLASSES)) model.add(Activation('softmax')) size = 1 if optimizer.lower() == 'sgd': opt = SGD(lr=learning_rate * size, decay=weight_decay, momentum=momentum) elif optimizer.lower() == 'rmsprop': opt = RMSprop(lr=learning_rate * size, decay=weight_decay) else: opt = Adam(lr=learning_rate * size, decay=weight_decay) model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy']) return model def get_filenames(channel_name, channel): if channel_name in ['train', 'validation', 'eval']: return [os.path.join(channel, channel_name + '.tfrecords')] else: raise ValueError('Invalid data subset "%s"' % channel_name) def train_input_fn(): return _input(args.epochs, args.batch_size, args.train, 'train') def eval_input_fn(): return _input(args.epochs, args.batch_size, args.eval, 'eval') def validation_input_fn(): return _input(args.epochs, args.batch_size, args.validation, 'validation') def _input(epochs, batch_size, channel, channel_name): filenames = get_filenames(channel_name, channel) dataset = tf.data.TFRecordDataset(filenames) dataset = dataset.repeat(epochs) dataset = dataset.prefetch(10) # Parse records. dataset = dataset.map( _dataset_parser, num_parallel_calls=10) # Potentially shuffle records. if channel_name == 'train': # Ensure that the capacity is sufficiently large to provide good random # shuffling. buffer_size = int(NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN * 0.4) + 3 * batch_size dataset = dataset.shuffle(buffer_size=buffer_size) # Batch it up. dataset = dataset.batch(batch_size, drop_remainder=True) iterator = dataset.make_one_shot_iterator() image_batch, label_batch = iterator.get_next() return {INPUT_TENSOR_NAME: image_batch}, label_batch def _train_preprocess_fn(image): """Preprocess a single training image of layout [height, width, depth].""" # Resize the image to add four extra pixels on each side. image = tf.image.resize_image_with_crop_or_pad(image, HEIGHT + 8, WIDTH + 8) # Randomly crop a [HEIGHT, WIDTH] section of the image. image = tf.random_crop(image, [HEIGHT, WIDTH, DEPTH]) # Randomly flip the image horizontally. image = tf.image.random_flip_left_right(image) return image def _dataset_parser(value): """Parse a CIFAR-10 record from value.""" featdef = { 'image': tf.FixedLenFeature([], tf.string), 'label': tf.FixedLenFeature([], tf.int64), } example = tf.parse_single_example(value, featdef) image = tf.decode_raw(example['image'], tf.uint8) image.set_shape([DEPTH * HEIGHT * WIDTH]) # Reshape from [depth * height * width] to [depth, height, width]. image = tf.cast( tf.transpose(tf.reshape(image, [DEPTH, HEIGHT, WIDTH]), [1, 2, 0]), tf.float32) label = tf.cast(example['label'], tf.int32) image = _train_preprocess_fn(image) return image, tf.one_hot(label, NUM_CLASSES) def save_model(model, output): signature = tf.saved_model.signature_def_utils.predict_signature_def( inputs={'inputs': model.input}, outputs={'scores': model.output}) builder = tf.saved_model.builder.SavedModelBuilder(output+'/1/') builder.add_meta_graph_and_variables( sess=K.get_session(), tags=[tf.saved_model.tag_constants.SERVING], signature_def_map={"serving_default": signature}) builder.save() logging.info("Model successfully saved at: {}".format(output)) return def main(args): logging.info("getting data") train_dataset = train_input_fn() eval_dataset = eval_input_fn() validation_dataset = validation_input_fn() logging.info("configuring model") model = keras_model_fn(args.learning_rate, args.weight_decay, args.optimizer, args.momentum) callbacks = [] # ----- 수정 부분 (경로 수정) ----- callbacks.append(ModelCheckpoint(args.model_output_dir + '/checkpoint-{epoch}.h5')) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - _archiving/contribution/daekeun-ml/tensorflow-in-sagemaker-workshop/training_script/cifar10_keras_tensorboard_solution.py [34:211]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - sess = tf.Session() K.set_session(sess) logging.getLogger().setLevel(logging.INFO) tf.logging.set_verbosity(tf.logging.INFO) HEIGHT = 32 WIDTH = 32 DEPTH = 3 NUM_CLASSES = 10 NUM_DATA_BATCHES = 5 NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 10000 * NUM_DATA_BATCHES INPUT_TENSOR_NAME = 'inputs_input' # needs to match the name of the first layer + "_input" def keras_model_fn(learning_rate, weight_decay, optimizer, momentum): """keras_model_fn receives hyperparameters from the training job and returns a compiled keras model. The model will be transformed into a TensorFlow Estimator before training and it will be saved in a TensorFlow Serving SavedModel at the end of training. Args: hyperparameters: The hyperparameters passed to the SageMaker TrainingJob that runs your TensorFlow training script. Returns: A compiled Keras model """ model = Sequential() model.add(Conv2D(32, (3, 3), padding='same', name='inputs', input_shape=(HEIGHT, WIDTH, DEPTH))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(Conv2D(32, (3, 3))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.2)) model.add(Conv2D(64, (3, 3), padding='same')) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(Conv2D(64, (3, 3))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.3)) model.add(Conv2D(128, (3, 3), padding='same')) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(Conv2D(128, (3, 3))) model.add(BatchNormalization()) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.4)) model.add(Flatten()) model.add(Dense(512)) model.add(Activation('relu')) model.add(Dropout(0.5)) model.add(Dense(NUM_CLASSES)) model.add(Activation('softmax')) size = 1 if optimizer.lower() == 'sgd': opt = SGD(lr=learning_rate * size, decay=weight_decay, momentum=momentum) elif optimizer.lower() == 'rmsprop': opt = RMSprop(lr=learning_rate * size, decay=weight_decay) else: opt = Adam(lr=learning_rate * size, decay=weight_decay) model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy']) return model def get_filenames(channel_name, channel): if channel_name in ['train', 'validation', 'eval']: return [os.path.join(channel, channel_name + '.tfrecords')] else: raise ValueError('Invalid data subset "%s"' % channel_name) def train_input_fn(): return _input(args.epochs, args.batch_size, args.train, 'train') def eval_input_fn(): return _input(args.epochs, args.batch_size, args.eval, 'eval') def validation_input_fn(): return _input(args.epochs, args.batch_size, args.validation, 'validation') def _input(epochs, batch_size, channel, channel_name): filenames = get_filenames(channel_name, channel) dataset = tf.data.TFRecordDataset(filenames) dataset = dataset.repeat(epochs) dataset = dataset.prefetch(10) # Parse records. dataset = dataset.map( _dataset_parser, num_parallel_calls=10) # Potentially shuffle records. if channel_name == 'train': # Ensure that the capacity is sufficiently large to provide good random # shuffling. buffer_size = int(NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN * 0.4) + 3 * batch_size dataset = dataset.shuffle(buffer_size=buffer_size) # Batch it up. dataset = dataset.batch(batch_size, drop_remainder=True) iterator = dataset.make_one_shot_iterator() image_batch, label_batch = iterator.get_next() return {INPUT_TENSOR_NAME: image_batch}, label_batch def _train_preprocess_fn(image): """Preprocess a single training image of layout [height, width, depth].""" # Resize the image to add four extra pixels on each side. image = tf.image.resize_image_with_crop_or_pad(image, HEIGHT + 8, WIDTH + 8) # Randomly crop a [HEIGHT, WIDTH] section of the image. image = tf.random_crop(image, [HEIGHT, WIDTH, DEPTH]) # Randomly flip the image horizontally. image = tf.image.random_flip_left_right(image) return image def _dataset_parser(value): """Parse a CIFAR-10 record from value.""" featdef = { 'image': tf.FixedLenFeature([], tf.string), 'label': tf.FixedLenFeature([], tf.int64), } example = tf.parse_single_example(value, featdef) image = tf.decode_raw(example['image'], tf.uint8) image.set_shape([DEPTH * HEIGHT * WIDTH]) # Reshape from [depth * height * width] to [depth, height, width]. image = tf.cast( tf.transpose(tf.reshape(image, [DEPTH, HEIGHT, WIDTH]), [1, 2, 0]), tf.float32) label = tf.cast(example['label'], tf.int32) image = _train_preprocess_fn(image) return image, tf.one_hot(label, NUM_CLASSES) def save_model(model, output): signature = tf.saved_model.signature_def_utils.predict_signature_def( inputs={'inputs': model.input}, outputs={'scores': model.output}) builder = tf.saved_model.builder.SavedModelBuilder(output+'/1/') builder.add_meta_graph_and_variables( sess=K.get_session(), tags=[tf.saved_model.tag_constants.SERVING], signature_def_map={"serving_default": signature}) builder.save() logging.info("Model successfully saved at: {}".format(output)) return def main(args): logging.info("getting data") train_dataset = train_input_fn() eval_dataset = eval_input_fn() validation_dataset = validation_input_fn() logging.info("configuring model") model = keras_model_fn(args.learning_rate, args.weight_decay, args.optimizer, args.momentum) callbacks = [] # ----- 수정 부분 (경로 수정) ----- callbacks.append(ModelCheckpoint(args.model_output_dir + '/checkpoint-{epoch}.h5')) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -