Predictions for UpliftTreeClassifier differ by version 0.14.1 vs. 0.15

[cheecharron]

I am attempting to create a single-tree model of a 4-treatment dataset with a continuous outcome using the UpliftTreeClassifier. In version 0.15, this attempt generates an interpretable tree but likely unusable predictions for each treatment group: Whereas my observed outcomes range from -20 to 20, the outputted predictions appear to be bound by 0 and 1--I assume these correspond to probabilities for a binary outcome. However, in version 0.14.1, the outputted predictions tend to follow the range of the observed values, but there is no variance in predictions across cases (i.e., no subgroups). This also seems like an invalid solution. Is there any way to produce a simple single-tree model with more than 2 treatment groups and a continuous outcome?

[jeongyoonlee]

1. UpliftTreeClassifer is a classifier for the binary outcome variable. If you use a non-binary outcome, it won't work as intended.
2. Its prediction outputs are the treatment effect estimates of the treatments over the control, and their ranges can differ from the range of the outcome variable.
3. Currently, CausalTreeRegressor doesn't support multiple treatments (Support multiple treatments in CausalTreeRegressor #775). For a regressor with multiple treatments, can you use meta-learners instead?