Adam Li

It seems this test works when n_estimators is low, but not when I ramp to `50` or maybe `100`. This makes me think there is a sampling edge case error...

I think the issue is isolated to when dim_contiguous=False you can test this with a single tree iterated over 100-1000 different random seeds. When dim_contiguous=True , seems there is consistently...

This was first noticed while testing in https://github.com/neurodata/scikit-tree/pull/172

`from sktree.experimental import mutual_info_ksg` works for me. FYI this submodule is quite experimental and untested

Hmm hard to say, since I cannot reproduce said issue. Can you provide all the information required in a bug report? https://github.com/neurodata/scikit-tree/issues/new?assignees=&labels=bug&projects=&template=bug_report.md&title=

CCA = SVD + QR decomposiiton. So someone just needs to find a C++/Cython robust implementation we can link to and import within our Cython code. Maybe peruse scipy/numpy compiled-code...

As a first step, one can implement just a CCA computation in Cython and then write a `cpdef` wrapper to expose it in Python for testing. The second step would...

Some thoughts here in increasing order of complexity: ### Criterion The criterion we are trying to implement is a generalization that stems from the GRF paper: ``` # the general...

## Splitter When it comes to the splitter, the code is a bit more simple, as most of the computation is passed off to the Criterion object. However, this [splitter](https://github.com/py-why/EconML/blob/main/econml/tree/_splitter.pyx)...

The commit here should enable the partial_fit without really too much effort: https://github.com/neurodata/scikit-learn/pull/54/commits/368df7a381b14952f9cee477a7e81b3768dcd2d0