apply tft.TFTransformOutput.transformed_feature_spec to tf.data

[Debasish-Das-CK]

dataset = tf.contrib.data.make_batched_features_dataset(
file_pattern=transformed_examples,
batch_size=batch_size,
features=tf_transform_output.transformed_feature_spec(),
reader=tf.data.TFRecordDataset,
shuffle=True)
tft examples uses a contrib API which is not official. Can I use the following ?

dataset = tf.data.TFRecordDataset(filenames_list)
dataset = dataset.map(_parse_proto)
dataset.map(tf_transform_output.transformed_feature_spec())

[Debasish-Das-CK]

also I guess examples need to be updated with the official API...should I push a PR ?

[ravwojdyla]

@Debasish-Das-CK tf.contrib.data.make_batched_features_dataset is fully supported and recommended see https://www.youtube.com/watch?v=uIcqeP7MFH0.

[Debasish-Das-CK]

sounds good @ravwojdyla