Fixes for scikit-learn 0.24

.github/workflows/pythonapp.yml

@@ -70,7 +70,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        scikitlearn-version: [0.21.3, 0.22.2, 0.23.1]
+        scikitlearn-version: [0.21.3, 0.22.2, 0.23.2, 0.24.1]
         pandas-version: [0.25.3, 1.0.5]
 
     steps:

aikit/cross_validation.py

@@ -92,17 +92,18 @@ def create_cv(cv=3, y=None, classifier=False, shuffle=False, random_state=None):
     if cv is None:
         cv = 3
 
+    random_state_ = random_state if shuffle else None
     if isinstance(cv, sklearn.model_selection._split.numbers.Integral):
         if (
             classifier
             and (y is not None)
             and (sklearn.model_selection._split.type_of_target(y) in ("binary", "multiclass"))
         ):
 
-            return sklearn.model_selection.StratifiedKFold(cv, shuffle=shuffle, random_state=random_state)
+            return sklearn.model_selection.StratifiedKFold(cv, shuffle=shuffle, random_state=random_state_)
 
         else:
-            return sklearn.model_selection.KFold(cv, shuffle=shuffle, random_state=random_state)
+            return sklearn.model_selection.KFold(cv, shuffle=shuffle, random_state=random_state_)
 
     if not hasattr(cv, "split") or isinstance(cv, str):
         if not isinstance(cv, sklearn.model_selection._split.Iterable) or isinstance(cv, str):

aikit/models/rotation_forest.py

@@ -11,9 +11,16 @@
 from sklearn.exceptions import NotFittedError
 from sklearn.base import ClassifierMixin, BaseEstimator, TransformerMixin, RegressorMixin
 
-from sklearn.ensemble.forest import ForestClassifier, ForestRegressor
+try:
+    from sklearn.ensemble.forest import ForestClassifier, ForestRegressor
+except ImportError:
+    from sklearn.ensemble._forest import ForestClassifier, ForestRegressor
+
 from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
-from sklearn.tree.tree import DTYPE
+try:
+    from sklearn.tree.tree import DTYPE
+except ImportError:
+    from sklearn.tree._tree import DTYPE
 
 
 from sklearn.preprocessing import StandardScaler
@@ -182,7 +189,6 @@ def __init__(
         min_impurity_decrease=0.0,
         min_impurity_split=None,
         class_weight=None,
-        presort=False,
         pca_bootstrap=False,
         pca_max_nb_groups=0.25,
         pca_max_group_size=0.05,
@@ -200,7 +206,6 @@ def __init__(
         self.min_impurity_decrease = min_impurity_decrease
         self.min_impurity_split = min_impurity_split
         self.class_weight = class_weight
-        self.presort = presort
 
         self.pca_bootstrap = pca_bootstrap
         self.pca_max_nb_groups = pca_max_nb_groups
@@ -237,7 +242,6 @@ def fit(self, X, y, sample_weight=None, check_input=True, X_idx_sorted=None):
             random_state=self.random_state,
             min_impurity_decrease=self.min_impurity_decrease,
             min_impurity_split=self.min_impurity_split,
-            presort=self.presort,
         )
 
         # 3) Apply group PCA
@@ -329,7 +333,6 @@ def __init__(
         max_leaf_nodes=None,
         min_impurity_decrease=0.0,
         min_impurity_split=None,
-        presort=False,
         pca_bootstrap=False,
         pca_max_nb_groups=0.25,
         pca_max_group_size=0.05,
@@ -346,7 +349,6 @@ def __init__(
         self.max_leaf_nodes = max_leaf_nodes
         self.min_impurity_decrease = min_impurity_decrease
         self.min_impurity_split = min_impurity_split
-        self.presort = presort
 
         self.pca_bootstrap = pca_bootstrap
         self.pca_max_nb_groups = pca_max_nb_groups
@@ -379,7 +381,6 @@ def fit(self, X, y, sample_weight=None, check_input=True, X_idx_sorted=None):
             random_state=self.random_state,
             min_impurity_decrease=self.min_impurity_decrease,
             min_impurity_split=self.min_impurity_split,
-            presort=self.presort,
         )
 
         # 3) Apply group PCA

aikit/scorer.py

@@ -8,23 +8,32 @@
 logger = logging.getLogger(__name__)
 
 import sklearn.metrics
-from sklearn.metrics.regression import _check_reg_targets, r2_score
+
+try:
+    from sklearn.metrics.regression import _check_reg_targets, r2_score
+except ImportError:
+    from sklearn.metrics import r2_score
+    from sklearn.metrics._regression import _check_reg_targets
+
 from sklearn.metrics import silhouette_score, davies_bouldin_score
 try:
     from sklearn.metrics import calinski_harabasz_score
 except ImportError:
     from sklearn.metrics import calinski_harabaz_score
     calinski_harabasz_score = calinski_harabaz_score
 
-
-from sklearn.metrics.scorer import SCORERS, _BaseScorer, type_of_target
+try:
+    from sklearn.metrics.scorer import SCORERS, _BaseScorer, type_of_target
+except ImportError:
+    from sklearn.metrics._scorer import SCORERS, _BaseScorer, type_of_target
 
 
 import numpy as np
 import pandas as pd
 
 from functools import partial
 
+
 class log_loss_scorer_patched(object):
     """ Log Loss scorer, correcting a small issue in sklearn (labels not used) """
 

aikit/tools/helper_functions.py

@@ -24,7 +24,12 @@
 from io import StringIO
 import hashlib
 
-from sklearn.utils import check_random_state, safe_indexing
+from sklearn.utils import check_random_state
+try:
+    from sklearn.utils import safe_indexing
+except ImportError:
+    from sklearn.utils import _safe_indexing
+    safe_indexing = _safe_indexing
 
 from aikit.tools.json_helper import SpecialJSONEncoder
 

aikit/transformers/base.py

@@ -17,8 +17,13 @@
 
 from sklearn.base import BaseEstimator, TransformerMixin, RegressorMixin
 from sklearn.exceptions import NotFittedError
-from sklearn.metrics.scorer import _BaseScorer, _PredictScorer
 
+try:
+    from sklearn.metrics.scorer import _BaseScorer, _PredictScorer
+    import sklearn.metrics.scorer as sk_scorer
+except ImportError:
+    from sklearn.metrics._scorer import _BaseScorer, _PredictScorer
+    import sklearn.metrics._scorer as sk_scorer
 
 from sklearn.utils import check_random_state
 from sklearn.utils.multiclass import type_of_target
@@ -33,7 +38,6 @@
 
 from sklearn.utils.metaestimators import if_delegate_has_method
 from sklearn.cluster import KMeans
-import sklearn.metrics.scorer
 
 # from aikit.helper_functions import is_user
 from aikit.enums import DataTypes
@@ -314,6 +318,7 @@ def __init__(
         self.n_components = n_components
         self.selector_type = selector_type
         self.component_selection = component_selection
+        self.random_state=random_state
         self.model_params = model_params
         self.columns_to_use = columns_to_use
         self.regex_match = regex_match
@@ -339,6 +344,7 @@ def _get_model(self, X, y=None):
             component_selection=self.component_selection,
             selector_type=self.selector_type,
             model_params=self.model_params,
+            random_state=self.random_state
         )
 
 
@@ -350,6 +356,7 @@ def __init__(
         n_components=0.5,
         selector_type="forest",
         component_selection="number",
+        random_state=None,
         model_params=None,
         columns_to_use="all",
         regex_match=False,
@@ -359,6 +366,7 @@ def __init__(
         self.n_components = n_components
         self.selector_type = selector_type
         self.component_selection = component_selection
+        self.random_state=random_state
         self.model_params = model_params
         self.columns_to_use = columns_to_use
         self.regex_match = regex_match
@@ -384,6 +392,7 @@ def _get_model(self, X, y=None):
             component_selection=self.component_selection,
             selector_type=self.selector_type,
             model_params=self.model_params,
+            random_state=self.random_state
         )
 
 
@@ -1008,15 +1017,15 @@ def _make_scorer(self, score_name):
         if isinstance(score_name, str):
 
             score_fun_dico = {
-                "explained_variance": sklearn.metrics.scorer.explained_variance_score,
-                "r2": sklearn.metrics.scorer.r2_score,
-                "neg_median_absolute_error": sklearn.metrics.scorer.median_absolute_error,
-                "neg_mean_absolute_error": sklearn.metrics.scorer.mean_absolute_error,
-                "neg_mean_squared_error": sklearn.metrics.scorer.mean_squared_error,
-                "neg_mean_squared_log_error": sklearn.metrics.scorer.mean_squared_log_error,
-                "median_absolute_error": sklearn.metrics.scorer.median_absolute_error,
-                "mean_absolute_error": sklearn.metrics.scorer.mean_absolute_error,
-                "mean_squared_error": sklearn.metrics.scorer.mean_squared_error,
+                "explained_variance": sk_scorer.explained_variance_score,
+                "r2": sk_scorer.r2_score,
+                "neg_median_absolute_error": sk_scorer.median_absolute_error,
+                "neg_mean_absolute_error": sk_scorer.mean_absolute_error,
+                "neg_mean_squared_error": sk_scorer.mean_squared_error,
+                "neg_mean_squared_log_error": sk_scorer.mean_squared_log_error,
+                "median_absolute_error": sk_scorer.median_absolute_error,
+                "mean_absolute_error": sk_scorer.mean_absolute_error,
+                "mean_squared_error": sk_scorer.mean_squared_error,
             }
 
             greater_is_better = {

requirements.txt

@@ -1,7 +1,7 @@
 scikit-learn>=0.20
 pandas>=0.23
 numpy
-scipy
+scipy<1.6.0                     # https://github.com/scikit-optimize/scikit-optimize/issues/981
 statsmodels
 lockfile
 decorator

tests/models/test_model_base.py

@@ -8,7 +8,7 @@
 import scipy.sparse as sps
 
 from sklearn.cluster import KMeans, DBSCAN, AgglomerativeClustering
-from sklearn.datasets.samples_generator import make_blobs
+from sklearn.datasets import make_blobs
 
 from aikit.models import DBSCANWrapper, KMeansWrapper, AgglomerativeClusteringWrapper
 

tests/models/test_rotation_forest.py

@@ -60,9 +60,7 @@ def test_GroupPCADecisionTreeClassifier():
     )
 
 
-pytest.mark.longtest
-
-
+@pytest.mark.longtest
 @pytest.mark.parametrize(
     "random_state, max_depth, criterion, pca_bootstrap",
     list(itertools.product(range(100), (None, 2, 5), ("gini", "entropy"), (True, False))),

tests/test_cross_validation.py

@@ -119,9 +119,13 @@ def test_fit_and_predict_transfrom():
     for train, test in cv.split(X, y):
 
         pt = DebugPassThrough()
-        predictions, _ = sklearn.model_selection._validation._fit_and_predict(
+        temp = sklearn.model_selection._validation._fit_and_predict(
             pt, X, y, train, test, verbose=1, fit_params=None, method="transform"
         )
+        if isinstance(temp, tuple):
+            predictions = temp[0]
+        else:
+            predictions = temp
 
         assert predictions.shape[0] == test.shape[0]
         assert predictions.shape[1] == X.shape[1]
@@ -138,9 +142,13 @@ def test_fit_and_predict_predict():
     for train, test in cv.split(X, y):
 
         logit = LogisticRegression()
-        predictions, _ = sklearn.model_selection._validation._fit_and_predict(
+        temp = sklearn.model_selection._validation._fit_and_predict(
             logit, X, y, train, test, verbose=1, fit_params=None, method="predict"
         )
+        if isinstance(temp, tuple):
+            predictions=temp[0]
+        else:
+            predictions=temp
 
         assert predictions.shape[0] == test.shape[0]
         assert len(predictions.shape) == 1
@@ -157,9 +165,13 @@ def test_fit_and_predict_predict_proba():
     for train, test in cv.split(X, y):
 
         logit = LogisticRegression()
-        predictions, _ = sklearn.model_selection._validation._fit_and_predict(
+        temp = sklearn.model_selection._validation._fit_and_predict(
             logit, X, y, train, test, verbose=1, fit_params=None, method="predict_proba"
         )
+        if isinstance(temp, tuple):
+            predictions=temp[0]
+        else:
+            predictions=temp
 
         assert predictions.shape[0] == test.shape[0]
         assert predictions.shape[1] == 2

tests/test_scorer.py

@@ -78,9 +78,13 @@ def test_avg_roc_auc_scorer_aikit():
 
     assert np.abs(cv_res1 - cv_res2).max() <= 10 ** (-5)
 
-    with pytest.raises(ValueError):
-        cross_val_score(logit, X, y, cv=cv, scoring="roc_auc")  # sklearn doesn't handle that
-
+    try:
+        res = cross_val_score(logit, X, y, cv=cv, scoring="roc_auc")  # sklearn doesn't handle that
+    except ValueError:
+        res = None
+    assert res is None or pd.isnull(res).all()
+    # sklearn <0.23 raise ValueError, sklearn >= 0.24 generates only 'nan'
+
     cv_res_aikit = cross_val_score(logit, X, 1 * (y == "AA"), cv=cv, scoring="avg_roc_auc")
     cv_res_sklearn = cross_val_score(logit, X, 1 * (y == "AA"), cv=cv, scoring="roc_auc")
 
@@ -107,8 +111,12 @@ def test_average_precision_scorer_aikit():
 
     assert np.abs(cv_res1 - cv_res2).max() <= 10 ** (-5)
 
-    with pytest.raises(ValueError):
-        cross_val_score(logit, X, y, cv=cv, scoring="average_precision")  # sklearn doesn't handle that
+    try:
+        res = cross_val_score(logit, X, y, cv=cv, scoring="average_precision")  # sklearn doesn't handle that
+    except ValueError:
+        res = None
+    assert res is None or pd.isnull(res).all()
+    # sklearn <0.23 raise ValueError, sklearn >= 0.24 generates only 'nan'
 
 
 def test_log_loss_patched_multioutput():