Computing Metrics for a Regression problem

[anujk3]

💡 Feature request

For all the sample notebooks given, we are computing the metrics where the outcome y is taken as a binary output, can we also do this for a regression problem and how it will change all the metrics and the corresponding curves?

Motivation

An example in the library for the regression problems will also help the users use sklift for similar problems

Additional context

Currently metrics assume y_val as a binary variable, what if its a regression problem? How to modify the metrics to not check for y_value as boolean and compute them?

[maks-sh]

Hi, @anujk3! Thanks for your feedback.

Indeed, the package supports only classification metrics now.

I think it is possible to start developing metrics for a regression problem with a Qini curve. More information can be found in section Continuous outcomes of the article Nicholas J Radcliffe. (2007). Using control groups to target on predicted lift: Building and assessing uplift model. Direct Marketing Analytics Journal, (3):14–21, 2007.