Test-Driven Machine Learning

Simulating real world situations

What I'd like to do now is take a detour to see what the performance of this algorithm looks like over time. We won't be test driving the simulation harness that I will use here. We will instead use this as an opportunity to visually identify the performance characteristics of two separate multi-armed bandit algorithms. Let's start introducing the different concepts from the code. The following code sets up our experimental world to be simulated. I'll instantiate BanditScenario and set up treatments A-C to have certain parameters that will provide guidance on how well each does in our experiment:

simulated_experiment = BanditScenario({ 'A': { 'conversion_rate': .05, 'order_average': 35.00 }, 'B':{ 'conversion_rate': ...