This is a repo to describe, test, and analyse the CUPED method using simulated experiments.

• dgp.py: Classes with data generating processes. Data is generated in the context of the driver-side of a taxi business but it can be easily put into another context. (The distributions used do not necessarily reflect the real-world).
• ate.py: Traditional and CUPED methods to estimate average treatment effect of global ratio variables. Also described as "non-user metrics" or "naive means" in different sources.
• explanation.md: Document (put together with the help of various sources) explaining the CUPED method and relevant material.
• sources.md: Document listing relevant sources.