What is the optimal batch size for a TensorFlow training?

See also: