Hands-On Q-Learning with Python

The testing process

In a typical instance of A/B testing, a marketer separates a group of users randomly into two groups and sends them each a version of an advertisement. The advertisement that gets the most clicks or conversions, depending on the performance metric used, is used the next time instead of the one that performed less well:

The preceding figure illustrates a typical comparison of three variations on an advertisement (A, B, and C). We can consider classic A/B testing to be using pure exploration with zero exploitation. We run some fixed number of trials, testing each option equally, and the one that performs best is the one that ...