Training Keras model with tf.data

mnist_tfdata.py

"""An example of how to use tf.Dataset in Keras Model"""
import tensorflow as tf   # only work from tensorflow==1.9.0-rc1 and after
 
_EPOCHS      = 5
_NUM_CLASSES = 10
_BATCH_SIZE  = 128

def training_pipeline():
  # #############
  # Load Dataset
  # #############
  (x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
  training_set = tfdata_generator(x_train, y_train, is_training=True, batch_size=_BATCH_SIZE)
  testing_set  = tfdata_generator(x_test, y_test, is_training=False, batch_size=_BATCH_SIZE)

  # #############
  # Train Model
  # #############
  model = keras_model()  # your keras model here
  model.compile('adam', 'categorical_crossentropy', metrics=['acc'])
  model.fit(
      training_set.make_one_shot_iterator(),
      steps_per_epoch=len(x_train) // _BATCH_SIZE,
      epochs=_EPOCHS,
      validation_data=testing_set.make_one_shot_iterator(),
      validation_steps=len(x_test) // _BATCH_SIZE,
      verbose = 1)


def tfdata_generator(images, labels, is_training, batch_size=128):
    '''Construct a data generator using tf.Dataset'''

    def preprocess_fn(image, label):
        '''A transformation function to preprocess raw data
        into trainable input. '''
        x = tf.reshape(tf.cast(image, tf.float32), (28, 28, 1))
        y = tf.one_hot(tf.cast(label, tf.uint8), _NUM_CLASSES)
        return x, y

    dataset = tf.data.Dataset.from_tensor_slices((images, labels))
    if is_training:
        dataset = dataset.shuffle(1000)  # depends on sample size

    # Transform and batch data at the same time
    dataset = dataset.apply(tf.contrib.data.map_and_batch(
        preprocess_fn, batch_size,
        num_parallel_batches=4,  # cpu cores
        drop_remainder=True if is_training else False))
    dataset = dataset.repeat()
    dataset = dataset.prefetch(tf.contrib.data.AUTOTUNE)

    return dataset



def keras_model():
    from tensorflow.keras.layers import Conv2D, MaxPool2D, Flatten, Dense, Dropout, Input

    inputs = Input(shape=(28, 28, 1))
    x = Conv2D(32, (3, 3),activation='relu', padding='valid')(inputs)
    x = MaxPool2D(pool_size=(2, 2))(x)
    x = Conv2D(64, (3, 3), activation='relu')(x)
    x = MaxPool2D(pool_size=(2, 2))(x)
    x = Flatten()(x)
    x = Dense(512, activation='relu')(x)
    x = Dropout(0.5)(x)
    outputs = Dense(_NUM_CLASSES, activation='softmax')(x)

    return tf.keras.Model(inputs, outputs)


if __name__ == '__main__':
  training_pipeline()

Hello, sorry if this is not directly related to this code sample. This is the most related post that I could find.
I would appreciate any help / reference.
I am trying to make tf.data.dataset work with tf.keras using the tf.estimator.
I get an error dimension when I am pushing the train_fn batch to the keras model (after converted it to estimator) . The code looks like that:

`def train_input_fn(batch_size=1):
"""An input function for training"""
print("train_input_fn: start function")

train_dataset = tf.data.experimental.make_csv_dataset(CSV_PATH_TRAIN, batch_size=batch_size,label_name='label',
select_columns=["sample","label"])
print('train_input_fn: finished make_csv_dataset')
train_dataset = train_dataset.map(parse_features_vector)
print("train_input_fn: finished the map with pars_features_vector")
train_dataset = train_dataset.repeat().batch(batch_size)
print("train_input_fn: finished batch size. train_dataset is %s ", train_dataset)
return train_dataset

IMG_SHAPE = (160,160,3)
base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,
include_top = False,
weights = 'imagenet')

base_model.trainable = False
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=0.0001),
loss='binary_crossentropy',
metrics=['accuracy'])

estimator = tf.keras.estimator.model_to_estimator(keras_model = model, model_dir = './date')

#train_input_fn read a CSV of images, resize them and returns dataset batch
train_spec = tf.estimator.TrainSpec(input_fn=train_input_fn, max_steps=20)

#eval_input_fn read a CSV of images, resize them and returns dataset batch of one sample
eval_spec = tf.estimator.EvalSpec(eval_input_fn)

tf.estimator.train_and_evaluate(estimator, train_spec=train_spec, eval_spec=eval_spec)`

the log is:
train_input_fn: finished batch size. train_dataset is %s <BatchDataset shapes: ({mobilenetv2_1.00_160_input: (None, 1, 160, 160, 3)}, (None, 1)), types: ({mobilenetv2_1.00_160_input: tf.float32}, tf.int32)>

and the error is:
ValueError: Input 0 of layer Conv1_pad is incompatible with the layer: expected ndim=4, found ndim=5. Full shape received: [None, 1, 160, 160, 3]

Many thanks for any help,
eilalan

I noticed the delayed training issue and noticed a big speed improvement in tensorflow-gpu-1.13 on my machine when I run tf.enable_eager_execution() before running the code.

I found tf1.14 still has this issue.
Is there a solution, anyone find it?

replace tf.contrib.data -> tf.data.experimental
remove .make_one_shot_iterator
and this runs

tf.__version__
 '2.2.0'

Python 3.8.3