Training, predicting, and preparing the drawing classification model

It's pretty straightforward to train the model but a little tricky to prepare the model for mobile deployment. Before we can start training, first make sure you already have the TensorFlow model repo (https://github.com/tensorflow/models) cloned in your TensorFlow root directory, as we did in the previous two chapters. Then download the drawing classification training dataset at http://download.tensorflow.org/data/quickdraw_tutorial_dataset_v1.tar.gz, which is about 1.1 GB, create a new folder called rnn_tutorial_data, and unzip the dataset tar.gz file to it. You'll see 10 training TFRecord files and 10 evaluation TFRecord files, as well as two files with the .classes extension, which have the same content and are just plain text for the 345 categories that the dataset can be used to classify, such...