The documents in this unit dive into the details of writing TensorFlow code. This section begins with the following guides, each of which explain a particular aspect of TensorFlow:
- Variables: Creation, Initialization, Saving, and Loading, which details the mechanics of TensorFlow Variables.
- Tensor Ranks, Shapes, and Types, which explains Tensor rank (the number of dimensions), shape (the size of each dimension), and datatypes.
- Sharing Variables, which explains how to share and manage large sets of variables when building complex models.
- Threading and Queues, which explains TensorFlow's rich queuing system.
- Reading Data, which documents three different mechanisms for getting data into a TensorFlow program.
The following guide is helpful when training a complex model over multiple days:
- Supervisor: Training Helper for Days-Long Trainings, which explains how to gracefully handle system crashes during a lengthy training session.
TensorFlow provides a debugger named
tfdbg, which is documented in the
following two guides:
- TensorFlow Debugger (tfdbg) Command-Line-Interface Tutorial: MNIST,
which walks you through the use of
tfdbgwithin an application written in the low-level TensorFlow API.
- How to Use TensorFlow Debugger (tfdbg) with tf.contrib.learn,
which demonstrates how to use
tfdbgwithin the Estimators API.
MetaGraph consists of both a computational graph and its associated
MetaGraph contains the information required to continue
training, perform evaluation, or run inference on a previously
trained graph. The following guide details
To learn about the TensorFlow versioning scheme, consult the following two guides:
- TensorFlow Version Semantics, which explains TensorFlow's versioning nomenclature and compatibility rules.
- TensorFlow Data Versioning: GraphDefs and Checkpoints, which explains how TensorFlow adds versioning information to computational graphs and checkpoints in order to support compatibility across versions.
We conclude this section with a FAQ about TensorFlow programming: