TST use global_random_seed in sklearn/cluster/tests/test_spectral.py #24802
Conversation
|
Hi @irene000 , thanks for your pull request! This will fix the lint issue and trigger all the other checks. Thanks! |
…/irene000/scikit-learn into test_spectral_global_random_seed
…/irene000/scikit-learn into test_spectral_global_random_seed
test_spectral_clustering test_spectral_clustering_sparse test_precomputed_nearest_neighbors_filtering test_affinities test_cluster_qr test_cluster_qr_permutation_invariance test_discretize test_spectral_clustering_with_arpack_amg_solvers test_n_components
test_spectral_clustering test_spectral_clustering_sparse test_precomputed_nearest_neighbors_filtering test_affinities test_cluster_qr test_cluster_qr_permutation_invariance test_discretize test_spectral_clustering_with_arpack_amg_solvers test_n_components
|
For the record I slightly modified the dataset in one of the tests because it was failing for a single seed and after a deep dive into the algorithm it turns out that there's inherent randomness in scipy's I suspect that it's just the result of rouding errors that accumulate differently in different calls because of the multi-threaded underlying BLAS/LACPACK calls, in which case it's kind of expected and there's nothing to do about it. |
Reference Issues/PRs
#22827
What does this implement/fix? Explain your changes.
Use global_random_seed in tests of test_spectral.py
Any other comments?