Reporting a bug

• [x]

Problem: the approximate method can still be slow for many trees
catboost version: master
Operating System: ubuntu 18.04
CPU: i9
GPU: RTX2080

Would be good to be able to specify how many trees to use for shapley. The model.predict and prediction_type versions allow this. lgbm/xgb allow this.

Is your feature request related to a problem? Please describe.
It might be useful to have a singular clean and performant way to check if all the columns of a dataframe are of the same dtype, such as a DataFrame property _is_homogeneous. This comes up in a lot of places, such as where we might want to dispatch to a cupy matrix implementation (Transpose, some row wise reductions I believe

Current implementation of join can be improved by performing the operation in a single call to the backend kernel instead of multiple calls.

This is a fairly easy kernel and may be a good issue for someone getting to know CUDA/ArrayFire internals. Ping me if you want additional info.

Names map and input are exchanged mistakenly. By sense of Preconditions paragraph they have to be exchanged I suppose, because there is no problem when map and result coincide (in current context).

The marker no_bad_cuml_array_check was necessary in 0.16 to avoid asserting on bad uses of CumlArray in specific tests. As of 0.17, this is no longer necessary.

This marker should be removed from pytest.ini as well as any tests that used it. A quick search shows this is used in 2 tests in test_incrementa_pca.py

I often use -v just to see that something is going on, but a progress bar (enabled by default) would serve the same purpose and be more concise.

We can just factor out the code from futhark bench for this.

Thank you for this fantastic work!

Could it be possible the fit_transform() method returns the KL divergence of the run?

Thx!