The Simple Practical Path to Machine Learning Capability: Models with Learned Parameters

In part one, we showed how the machine learning process is like the scientific thinking process, and in part two, we introduced a benchmark task and showed how to get your machine learning system up and running with a simple nearest neighbors model. Time to crank it up! In this post, we continue the scientific thinking process by extending our simple starting model into increasingly powerful models. We'll show how to implement a particularly useful kind of model in Tensorflow. Shortcutting this process is a common failure mode--ye've been warned! But familiarity with common patterns of errors, diagnoses, and solutions allow experts to zip through the iterations. From the perspective of someone who is learning, it might look like the expert is skipping steps and going straight to the complicated stuff.