search

wayfair / pylift

Uplift modeling package.
http://pylift.readthedocs.io
BSD 2-Clause "Simplified" License
368 stars 76 forks source link

multiple treatment groups #24

Closed anaveenan closed 5 years ago

anaveenan commented 5 years ago

I have experiment data with multiple treatment groups. How do I include this model ?

rsyi commented 5 years ago

You'd have to make multiple pylift.TransformedOutcome objects as well, unfortunately.

pylift seeks to maximize the lift between a treatment and another treatment (the control). We generally want to do this because it'll find splits that are more optimal for maximal lift than splitting by purity.

It's tricky to do this with multiple treatments, as you'd have to decide if you want to find the splits that maximize the difference between, for example, treatment A and control, treatment B and control, or between treatment A and treatment B. If just the first two (as should generally be the case), you can make two different models and choose the lift values that are the highest between them.