FIX support read only view for tree input

[Batalex]

Reference Issues/PRs

Fixes #15851.

What does this implement/fix? Explain your changes.

According to https://cython.readthedocs.io/en/latest/src/userguide/memoryviews.html#read-only-views

Since Cython 0.28, the memoryview item type can be declared as const to support read-only buffers as input

This should be fine with scikit-learn since the minimal Cython version supported is 28.5. We could also check the input array for the writeable flag and make a copy instead of a view if needed but I figured this solution would be better.

Any other comments?

The benchmarks now work as expected.

python benchmarks\bench_mnist.py --classifiers RandomForest ExtraTrees CART
[...]
Classification performance:
===========================
Classifier               train-time   test-time   error-rate
------------------------------------------------------------
ExtraTrees                   45.16s       0.45s       0.0294
RandomForest                 40.51s       0.39s       0.0295
CART                         21.61s       0.06s       0.1221

2020 Paris Sprint

[rth]

We should do the same in _decision_path_dense otherwise this sounds good to me.

Please add a unit tests that check that trees work with read-only array as input (you can use create_memmap_backed_data to make one).

Also please add a what's new entry to doc/whats_new/v0.23.rst indicating that the corresponding tree methods can now be used with joblib based multiprocessing (or @jeremiedbb might suggest a better formulation).

[Batalex]

I believe we can remove the test test_decision_tree_memmap https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/tree/tests/test_tree.py#L1826. It should be covered by the new ones and it will remove an import.

[rth]

Thanks @Batalex, LGTM.

I pushed a small fix to improve the what's new entry given new information from #15851 (comment) and also removed test_decision_tree_memmap as you suggested since it is indeed redundant.

Will merge on green CI.

[rth]

The list of affected estimators is obtained in #16359