Adam Li

One idea is that we add the loop through `n_samples`. For each feature that is involved in the projection vector, we keep track of the min/max for that column. This...

I think we can do this for SPORF, but not for general MORF. Because if we move around the feature indices (i.e. columns), then it breaks the "structure" of the...

In the PR addressing this, you can add two examples: 1. An example showing off the unsupervised tree functionality and comparing it to say k-means for the iris dataset. Very...

Will be closed by: https://github.com/neurodata/scikit-learn/pull/46

A benchmarking done using cc18's openml dataset with categorical features would be nice: https://github.com/scikit-learn/scikit-learn/pull/12866#issuecomment-455350207 Basically run sklearn w/o categorical support and one-hot encoding vs w/ categorical support - track runtime...

Ideally this causal tree class allows all the functionality that econml currently supports: - grf (probably just for sake of continuity) - doubly robust trees - dml trees - orf...

IIUC, GRF are just regular trees with: - honesty - some fancy criterion that solves a moment equation I don't think we really need to implement this as a high...

Do you know if the other types of trees need to use more exotic criterions? E.g. in econml, the 'het' and 'mse' are these moment equation solvers. I guess for...

Okay that's good to know. The parameter would just be the normal `criterion` keyword argument, which can be 'gini', 'poisson', etc.

Re 'het' and the 'mse' (note this mse is not the same as the MSE currently in sklearn, so we should probably call it something else...) criterion in econml, we'll...