Release 0.22rc3 #15715
Release 0.22rc3 #15715
Conversation
* documentation for random_state in forests * move note to parameter * same for RandomForestRegressor * add doc for ExtraTreesRegressor and ExtraTreesClassifier * skip line * lint * move note back to where it was * add Glossary in RandomForestRegressor * adding description for RandomTreesEmbedding * small fix * correct description for RandomTreesEmbedding
…1514) * First cut at basic single linkage internals * Refer to correct dist_metrics package * Add csgraph sparse implementation for single linkage * Add fast labelling/conversion from MST to single linkage tree; remove uneeded single_linkage.pyx file. * Ensure existing tests cover single linkage * Name cingle linkage labelling correctly. * Iterating toward correct solution. Still have to get n_clusters, compute_full_tree=False working * Get n_components correct. * Update docstrings. * Fix the parents array when we don't get the "full tree" * Add single linkage to agglomerative clustering example. * Add single linkage to digits agglomerative clustering example. * Update documentation to reflect the addition of single linkage. * Update documentation to reflect the addition of single linkage. * Pep8 fix for class declaration in cython * Fix heading in clustering docs * Update the digits clustering text to reflect the new reality. * Provide a more complete comparison of the different linkage methods, highlighting the relative strengths and weaknesses. * We don't need connectivity here, and we can ignore issues with warnings for spectral clustering. * Add an explicit test that single linkage successfully works on examples it should perform well on. * Update docs with a more complete comparison on linkage methods (scale to be determined?) * List formatting in example linkage comparison. * Flake8 fixes. * Flake8 fixes. * More Flake8 fixes. * Fix agglomerative plot example with correct subplot spec * Explicitly test linkages (including single) produce results identical to scipy.cluster.hierarchical * Fix comment on why we sort (consistency) * Make dense single linkage faster * Add docstring to new mst-linkage-core computations. * Add a test that new single linkage code matches scipy * Ensure we only attemtp this for metrics Jake implemented. * Per amueller; it's a long paper, ref the figure. * Clean up a few things. * Too many blank lines for flake8 * Bad scipy slink input * Flake8 fixes * Clean up cython a little; fix typo/carryover * Convert memoryview to numpy array on return * Just convert to the correct dtype * Update sklearn/cluster/_hierarchical.pyx Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Update sklearn/cluster/_hierarchical.pyx Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Update sklearn/cluster/_hierarchical.pyx Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Update sklearn/cluster/tests/test_hierarchical.py Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Fixes as per @NicolasHug suggestions. * Update renaming of params in test_hierarchical * Relative import? * Ah, it got renamed in master... * A bad merge on my part. * In principle this is in sklearn.neighbors now... * No; not that way... * Declare dim before use. * Update sklearn/cluster/tests/test_hierarchical.py Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Remaining fixes per Nicolas Hug. * Update sklearn/cluster/tests/test_hierarchical.py Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Fix flake8 issues. * Switch from stable to mergesort per jnotham * Update sklearn/cluster/_hierarchical.py Co-Authored-By: Nicolas Hug <contact@nicolas-hug.com> * Skip checks that are already validated. * Update docstring per Gael's suggestion * Add a benchmark script for agglomerative clustering * Fix some flake8 issues * No flake8 on the one line * Update parameters and output for benchmark hierarchical * Switch to 2D plotting for hierarchical benchmark * Wrong colormap name * Formatting fpr bench hierarchical * Add an item to WhatsNew
… RadiusNeighborsRegressor (scikit-learn#15688)
* whats_new cleanup * fix indent * reorder hgbt ones * add Guillaume to contributors and remove unnecessary entry
|
I'd usually exclude something like #11514 after a feature freeze unless it was very clear that consensus was to release it in a hurry. |
If we do that, (I am +0 for not including it to respect our agreed upon process), we need to change the what's new entry to move it to 0.23 in master. |
|
I think we can ignore the circleci failures since they seem to be absent if we do a Although it does make me think if we should fix our script. These seem false positives to me. |
|
Yea the script needs to be adjusted to either not error on images from the api docs or run the examples for them. |
* include all pxd files in package * _libsvm _liblinear to pxi + add a test * simplify stuff + use pathlib * cln * more cln
|
I think #15711 is a query about a non-released feature that should be
resolved ideally before making it hard to change. But it only needs a
decision not to change if we're happy with that.
|
|
@adrinjalali , @thomasjpfan , yes this is clearly a false positive but the script merged in #15680 should work for all the failing cases in this PR. not as expected. |
|
This was dropped, so it is using the old |
|
All green. Shall we merge for me to tag and make the wheels? |
|
Do we expect any further changes before final? |
|
Perhaps add c4733f4 and 63cd600. |
|
Maybe we can skip the rc3 and release directly? |
|
What are the HGBDT issues @qinhanmin2014 ? |
|
|
Intention of rc3 is partially to test a py3.8 wheel, but yes it's not
strictly necessary
|
|
Ready to merge, tag and proceed with release as far as I'm concerned |
I've included the doc changes, and the ones which were suggested to be included.
The only one I was a bit surprised about is #11514 which I've included since it touches the whats_new, but it's not a very minor change.
It'd be also nice to have #15626 here.