forked from intel/scikit-learn-intelex
-
Notifications
You must be signed in to change notification settings - Fork 0
/
deselected_tests.yaml
executable file
·96 lines (74 loc) · 4.82 KB
/
deselected_tests.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
# This file lists node ids (in pytest sense) of sklearn tests that
# are to be deselected during test discovery step.
#
# Deselection can be predicated on the version of scikit-learn used.
# Use - node_id cond, or - node_id cond1,cond2 where cond is OPver.
# Supported OPs are >=, <=, ==, !=, >, <
# For example,
# - tests/test_isotonic.py::test_permutation_invariance >0.18,<=0.19
# will exclude deselection in versions 0.18.1, and 0.18.2 only
deselected_tests:
# Deselecting 5 SVC related tests where duplicate samples end up being support vectors
# See: https://github.com/scikit-learn/scikit-learn/issues/12738
- svm/tests/test_svm.py::test_svc_clone_with_callable_kernel
- svm/tests/test_svm.py::test_precomputed
- svm/tests/test_sparse.py::test_sparse_realdata
- svm/tests/test_sparse.py::test_svc_iris
# Passed from DAAL2020.4. Need to delete after release.
- ensemble/tests/test_bagging.py::test_sparse_classification
- svm/tests/test_sparse.py::test_unsorted_indices
# Bitwise comparison of probabilities using a print.
- metrics/tests/test_classification.py
# Max absolute difference: 0.04 for rocauc, and 0.01 for precision_recallq
- metrics/tests/test_ranking.py::test_roc_curve_hard
- metrics/tests/test_ranking.py::test_precision_recall_curve
- model_selection/tests/test_search.py::test_search_cv_results_rank_tie_breaking
# test_k_means_fit_predict is known to sporadically fail for float32 inputs in multithreaded runs
# See: https://github.com/IntelPython/daal4py/issues/25
- cluster/tests/test_k_means.py::test_k_means_fit_predict
# test_non_uniform_strategies fails due to differences in handling of vacuous clusters after update
# See https://github.com/IntelPython/daal4py/issues/69
- preprocessing/tests/test_discretization.py::test_nonuniform_strategies >=0.20.3
- cluster/tests/test_k_means.py::test_relocated_clusters
- cluster/tests/test_k_means.py::test_kmeans_relocated_clusters
# DAAL does not check convergence for tol == 0.0 for ease of benchmarking
- cluster/tests/test_k_means.py::test_kmeans_convergence
- cluster/tests/test_k_means.py::test_kmeans_verbose
# For Newton-CG solver, solution computed in float32 disagrees with that of float64 a little more than
# the test expects, see https://github.com/scikit-learn/scikit-learn/pull/13645
- linear_model/tests/test_logistic.py::test_dtype_match
# _hist_gradient_boosting tests (added in 0.21.0) times out in CircleCI
- ensemble/_hist_gradient_boosting/tests >=0.21.0,<0.21.1
# https://github.com/scikit-learn/scikit-learn/pull/15857
- ensemble/tests/test_stacking.py::test_stacking_cv_influence >=0.22
# This fails on certain platforms. Weighted data do not go through DAAL,
# unweighted do. Since convergence is not accomplished (comment in te test
# suggests that), coefficients are slightly different, resulting in 1 prediction
# disagreement.
- ensemble/tests/test_stacking.py::test_stacking_with_sample_weight >=0.22.1
# test is unwarranted, but upstream refuses to change it
# https://github.com/scikit-learn/scikit-learn/pull/15856
- svm/tests/test_svm.py::test_svm_equivalence_sample_weight_C >=0.22
#
- tests/test_pipeline.py::test_pipeline_memory
# docstrings do not document daal_model_ attribute set by fit for
# LinearRegression, Ridge and SVC
- tests/test_docstring_parameters.py::test_fit_docstring_attributes
# Insufficient accuracy of "coefs" and "intercept" in Elastic Net for multi-target problem
# https://github.com/oneapi-src/oneDAL/issues/494
- linear_model/tests/test_coordinate_descent.py::test_enet_multitarget >=0.21
# Insufficient accuracy of objective function in Elastic Net in case warm_start
# https://github.com/oneapi-src/oneDAL/issues/495
- linear_model/tests/test_coordinate_descent.py::test_warm_start_convergence_with_regularizer_decrement >=0.21
# DAAL doesn't support sample_weight (back to Sklearn), insufficient accuracy (similar to previous cases)
- linear_model/tests/test_coordinate_descent.py::test_enet_sample_weight_consistency >=0.23
# We have difference (Max absolute difference: 1.97215226e-31) in computing of log_proba with np.log(y_proba)
# Looks like using IDP NumPy is the cause for the failure due to use of more aggressive compiler optimization when compile NumPy UFunc loops
- neural_network/tests/test_mlp.py::test_predict_proba_multilabel
# On small datasets, the regression coefficients for multi-target problem differ from scikit-learn. Coefficients matches for first label only.
# For big data the coefficients are close.
# See: https://github.com/IntelPython/daal4py/issues/275
- linear_model/tests/test_ridge.py::test_ridge_cv_individual_penalties
# Different number of iterations because of instability of kmeans
# https://github.com/IntelPython/daal4py/issues/277
- cluster/tests/test_k_means.py::test_kmeans_elkan_results