def parse_tfos(example_proto):
    feature_def = {"label": tf.io.FixedLenFeature(1, tf.int64),
                   "image": tf.io.FixedLenFeature(28 * 28 * 1, tf.int64)}
    features = tf.io.parse_single_example(example_proto, feature_def)
    image = tf.cast(features['image'], tf.float32) / 255
    image = tf.reshape(image, (28, 28, 1))
    label = tf.cast(features['label'], tf.int32)
    return (image, label)

image_pattern = ctx.absolute_path(args.images_labels)
print("image_pattern is {0}".format(image_pattern))
ds = tf.data.Dataset.list_files(image_pattern)
ds = ds.repeat(args.epochs).shuffle(BUFFER_SIZE)
ds = ds.interleave(lambda x: tf.data.TFRecordDataset(x, compression_type='GZIP'))
train_datasets_unbatched = ds.map(parse_tfos)

def build_and_compile_cnn_model():
    model = tf.keras.Sequential([
        tf.keras.layers.Conv2D(32, 3, activation='relu', input_shape=(28, 28, 1)),
        tf.keras.layers.MaxPooling2D(),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(64, activation='relu'),
        tf.keras.layers.Dense(10, activation='softmax')
    ])
    model.compile(
        loss=tf.keras.losses.sparse_categorical_crossentropy,
        optimizer=tf.keras.optimizers.SGD(learning_rate=0.001),
        metrics=['accuracy'])
    return model

# single node
# single_worker_model = build_and_compile_cnn_model()
# single_worker_model.fit(x=train_datasets, epochs=3)

# Here the batch size scales up by number of workers since
# `tf.data.Dataset.batch` expects the global batch size. Previously we used 64,
# and now this becomes 128.
GLOBAL_BATCH_SIZE = BATCH_SIZE * NUM_WORKERS
train_datasets = train_datasets_unbatched.batch(GLOBAL_BATCH_SIZE)

# this fails
# callbacks = [tf.keras.callbacks.ModelCheckpoint(filepath=args.model_dir)]
tf.io.gfile.makedirs(args.model_dir)
filepath = args.model_dir + "/weights-{epoch:04d}"
callbacks = [tf.keras.callbacks.ModelCheckpoint(filepath=filepath, verbose=1, save_weights_only=True)]

# Note: if you part files have an uneven number of records, you may see an "Out of Range" exception
# at less than the expected number of steps_per_epoch, because the executor with least amount of records will finish first.
steps_per_epoch = 60000 / GLOBAL_BATCH_SIZE

with strategy.scope():
    multi_worker_model = build_and_compile_cnn_model()
multi_worker_model.fit(x=train_datasets, epochs=args.epochs, steps_per_epoch=steps_per_epoch, callbacks=callbacks)

compat.export_saved_model(multi_worker_model, args.export_dir, ctx.job_name == 'chief')

notebook_ip = socket.gethostbyname(socket.gethostname())
notebook_port = "30202"
conf = SparkConf()
conf.setAppName("mnist_keras_tfrecord")
conf.set("spark.driver.host", notebook_ip)
conf.set("spark.driver.port", notebook_port)
sc = SparkContext(conf=conf)
executors = sc._conf.get("spark.executor.instances")
num_executors = int(executors) if executors is not None else 1

parser = argparse.ArgumentParser()
parser.add_argument("--batch_size", help="number of records per batch", type=int, default=64)
parser.add_argument("--buffer_size", help="size of shuffle buffer", type=int, default=10000)
parser.add_argument("--cluster_size", help="number of nodes in the cluster", type=int, default=num_executors)
parser.add_argument("--epochs", help="number of epochs", type=int, default=3)
parser.add_argument("--images_labels", help="HDFS path to MNIST image_label files in parallelized format")
parser.add_argument("--model_dir", help="path to save model/checkpoint", default="mnist_model")
parser.add_argument("--export_dir", help="path to export saved_model", default="mnist_export")
parser.add_argument("--tensorboard", help="launch tensorboard process", action="store_true")

args = parser.parse_args()
print("args:", args)

from tensorflowonspark import TFCluster

cluster = TFCluster.run(sc, main_fun, args, args.cluster_size, num_ps=0, tensorboard=args.tensorboard,
                        input_mode=TFCluster.InputMode.TENSORFLOW, master_node='chief')
cluster.shutdown()