Predicted vs actual uplift not possible anymore
steprandelli opened this issue · comments
Hi!
We are using uplift_by_percentile to compute predicted and actual uplift by decile of predicted uplift.
This is then used to compute MSRC as mentioned in Michel, R., Schnakenburg, I. and von Martens, T., 2019. Targeting Uplift. An Introduction to Net Scores/by René Michel, Igor Schnakenburg, Tobias von Martens.
Since PR 120 this is not possible due to check_is_binary(y_true) in uplift_by_percentile function.
We'd like to use this function passing to y_true a float with the prediction.
May we check y_true is a float instead of a boolean?
scikit-uplift/sklift/metrics/metrics.py
Line 653 in c9dd56a
Hello!
Sorry for the long answer.
Thank you for your feedback! This is a good idea.
Could you please contribute the code of Model Stability Rank Correlation metric?
In this PR you can remove this checker from uplift_by_percentile function.