halhorn

jjallaire

'''Train MNIST with tfrecords yielded from a TF Dataset

In order to run this example you should first run 'mnist_to_tfrecord.py'
which will download MNIST data and serialize it into 3 tfrecords files
(train.tfrecords, validation.tfrecords, and test.tfrecords).

This example demonstrates the use of TF Datasets wrapped by a generator
function. The example currently only works with a fork of keras that accepts
`workers=0` as an argument to fit_generator, etc. Passing `workers=0` results
in the generator function being run on the main thread (without this various