GitLab

1. `tf.contrib.data.batch_and_drop_remainder()` has been deprecated for a while now.
2. `Dataset.make_one_shot_iterator()` is no longer necessary in input functions, and avoiding it lead to better performance on TPUs. (It will also be removed in TF 2.0.)

Some changes specific to prediction. Removing traces of expected results, as this is just prediction.

Right now we don't have input data for prediction. So using top 10
entries of test data as input.

* Add new import paths to mnist_tpu.py

This adds new import paths to mnist_tpu.py so individuals can check out the git repository and run `python mnist_tpu.py` without finagling with python import search paths.

* Fix lint errors

* Add disable= to pylint instructions

* Remove parent path

* Update comments

* Fix MNIST to allow master=local to work again.

* Update mnist_tpu.py

* Glint everything

* Adding rcfile and pylinting

* Extra newline

* Few last lints

* Adding logging utils

* restore utils

* delete old file

* update inputs and docstrings

* make /official a python module

* remove /utils directory

* Update readme for python path setting

* Change readme texts

In TensorFlow 1.7, the TPUClusterResolver and the TPU RunConfig are changed
to reduce the amount of boilerplate required.

The long path doesn't work for TensorFlow 1.6 because the .python package under cluster_resolver doesn't exist.

The `sparse` version is more efficient anyway.

I'm returning the labels shape [1] instead of []
because tf.accuracy fails otherwise.

The `sparse` version is more efficient anyway.

I'm returning the labels shape [1] instead of []
because tf.accuracy fails otherwise.

This is a step towards merging the example in
https://github.com/tensorflow/tpu-demos/tree/master/cloud_tpu/models/mnist
with this repository, so we have a single model definition for
training across CPU/GPU/eager execution/TPU.

The change to dataset.py is so that the raw data can be read
from cloud storage systems (like GCS and S3).