I have experiment data with two treatment groups where treatment #2 is better than #1 across the entire population.
What's strange is that when I fit an UpliftTreeClassifier on treatment #2 and control data I get slightly different splits than when I fit an UpliftTreeClassifier on test #1, #2, and control. Even when looking at the uplift scores for the latter model, all the uplift scores are from treatment #2 vs control.
What could possibly lead to this happening?
I have experiment data with two treatment groups where treatment #2 is better than #1 across the entire population.
What's strange is that when I fit an UpliftTreeClassifier on treatment #2 and control data I get slightly different splits than when I fit an UpliftTreeClassifier on test #1, #2, and control. Even when looking at the uplift scores for the latter model, all the uplift scores are from treatment #2 vs control.
What could possibly lead to this happening?