Chapter 12

Developing Applications with Estimators

IN THIS CHAPTER

check Executing machine learning algorithms with estimators

check Defining features and feature columns

check Using pre-built estimators in practice

check Analyzing complex datasets with wide and deep learning

At a fundamental level, the process of using statistical regression for machine learning is a lot like the process of using neural networks (see Chapter 7): Load your data, train your model, and test the result. Unfortunately, the code needed to perform statistical regression in TensorFlow is quite different than the code needed to create neural networks.

To simplify development and testing, TensorFlow provides the Estimator framework. The tf.estimator package contains modules that analyze data through a common set of methods. For example, the estimator class that performs linear regression (LinearRegressor) has the same methods as the class that performs regression with deep neural networks (DNNRegressor).

You can take advantage of this commonality by coding your own estimators. That is, if you package your custom machine learning ...

Get TensorFlow For Dummies now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.