Update code for latest scipy and sphinx-gallery (#1551)

* Make meta-feature computation compatible with scipy 1.9

* Also fix doc for latest sphinx-gallery

Author: mfeurer

autosklearn/metalearning/metafeatures/metafeatures.py

@@ -530,11 +530,16 @@ def _calculate(self, X, y, logger, feat_type):
         kurts = []
         for i in range(X.shape[1]):
             if numerical[X.columns[i] if hasattr(X, "columns") else i]:
-                kurts.append(
-                    scipy.stats.kurtosis(
-                        X.iloc[:, i] if hasattr(X, "iloc") else X[:, i]
+                if np.isclose(
+                    np.var(X.iloc[:, i] if hasattr(X, "iloc") else X[:, i]), 0
+                ):
+                    kurts.append(0)
+                else:
+                    kurts.append(
+                        scipy.stats.kurtosis(
+                            X.iloc[:, i] if hasattr(X, "iloc") else X[:, i]
+                        )
                     )
-                )
         return kurts
 
     def _calculate_sparse(self, X, y, logger, feat_type):
@@ -548,7 +553,10 @@ def _calculate_sparse(self, X, y, logger, feat_type):
             if numerical[X.columns[i] if hasattr(X, "columns") else i]:
                 start = X_new.indptr[i]
                 stop = X_new.indptr[i + 1]
-                kurts.append(scipy.stats.kurtosis(X_new.data[start:stop]))
+                if np.isclose(np.var(X_new.data[start:stop]), 0):
+                    kurts.append(0)
+                else:
+                    kurts.append(scipy.stats.kurtosis(X_new.data[start:stop]))
         return kurts
 
 
@@ -594,9 +602,16 @@ def _calculate(self, X, y, logger, feat_type):
         skews = []
         for i in range(X.shape[1]):
             if numerical[X.columns[i] if hasattr(X, "columns") else i]:
-                skews.append(
-                    scipy.stats.skew(X.iloc[:, i] if hasattr(X, "iloc") else X[:, i])
-                )
+                if np.isclose(
+                    np.var(X.iloc[:, i] if hasattr(X, "iloc") else X[:, i]), 0
+                ):
+                    skews.append(0)
+                else:
+                    skews.append(
+                        scipy.stats.skew(
+                            X.iloc[:, i] if hasattr(X, "iloc") else X[:, i]
+                        )
+                    )
         return skews
 
     def _calculate_sparse(self, X, y, logger, feat_type):
@@ -610,7 +625,10 @@ def _calculate_sparse(self, X, y, logger, feat_type):
             if numerical[X.columns[i] if hasattr(X, "columns") else i]:
                 start = X_new.indptr[i]
                 stop = X_new.indptr[i + 1]
-                skews.append(scipy.stats.skew(X_new.data[start:stop]))
+                if np.isclose(np.var(X_new.data[start:stop]), 0):
+                    skews.append(0)
+                else:
+                    skews.append(scipy.stats.skew(X_new.data[start:stop]))
         return skews
 
 

doc/faq.rst

@@ -13,7 +13,7 @@ General
 
     We provide examples on using *auto-sklearn* for multiple use cases ranging from
     simple classification to advanced uses such as feature importance, parallel runs
-    and customization. They can be found in the :ref:`sphx_glr_examples`.
+    and customization. They can be found in the :ref:`examples`.
 
 .. collapse:: <b>What type of tasks can auto-sklearn tackle?</b>
 

doc/manual.rst

@@ -13,7 +13,7 @@ give short explanations (click the title to expand text), e.g.
 
     We provide examples on using *auto-sklearn* for multiple use cases ranging from
     simple classification to advanced uses such as feature importance, parallel runs
-    and customization. They can be found in the :ref:`sphx_glr_examples`.
+    and customization. They can be found in the :ref:`examples`.
 
 .. collapse:: <b>Material from talks and presentations</b>
 

doc/releases.rst

@@ -657,7 +657,7 @@ Version 0.4.0
   for the ensemble.
 * Adds an operating system check at import and installation time to make sure
   to not accidentaly run on a Windows machine.
-* New examples gallery using sphinx gallery: :ref:`sphx_glr_examples`
+* New examples gallery using sphinx gallery: :ref:`examples`
 * Safeguard Auto-sklearn against deleting directories it did not create (Issue
   `#317 <https://github.com/automl/auto-sklearn/issues/317>`_.
 

test/test_metalearning/pyMetaLearn/test_meta_features.py

@@ -148,17 +148,7 @@ def meta_train_data_transformed(request):
     if request.param == "numpy":
         return X_transformed, y, feat_type_transformed
     elif request.param == "pandas":
-        dtypes = {}
-        for key, value in feat_type.items():
-            if value == "categorical":
-                dtypes[key] = "category"
-            elif value == "numerical":
-                dtypes[key] = "float64"
-            elif value == "string":
-                dtypes[key] = "string"
-            else:
-                raise KeyError
-        X_transformed = pd.DataFrame(X_transformed).astype(dtypes)
+        X_transformed = pd.DataFrame(X_transformed)
         return X_transformed, y, feat_type_transformed
     else:
         raise ValueError(request.param)
@@ -881,10 +871,10 @@ def test_calculate_all_metafeatures_same_results_across_datatypes():
         "SkewnessMean": 1.47397188548894,
         "SkewnessMax": 29.916569235579203,
         "SkewnessMin": -29.916569235579203,
-        "KurtosisSTD": 153.0563504598898,
-        "KurtosisMean": 56.998860939761165,
+        "KurtosisSTD": 152.95700852863064,
+        "KurtosisMean": 57.258120199020425,
         "KurtosisMax": 893.0011148272025,
-        "KurtosisMin": -3.0,
+        "KurtosisMin": -1.998392219134577,
         "SymbolsSum": 49,
         "SymbolsSTD": 1.3679553264445183,
         "SymbolsMean": 1.8846153846153846,

test/test_metalearning/pyMetaLearn/test_meta_features_sparse.py

@@ -338,38 +338,36 @@ def test_skewnesses(sparse_data_transformed):
     mf = meta_features.helper_functions["Skewnesses"](
         X_transformed, y, logging.getLogger("Meta")
     )
-    print(mf.value)
-    print(fixture)
     np.testing.assert_allclose(mf.value, fixture)
 
 
 def test_kurtosisses(sparse_data_transformed):
     fixture = [
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
-        -3.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
         -1.1005836114255763,
         -1.1786325509475744,
         -1.23879983823279,
@@ -381,7 +379,6 @@ def test_kurtosisses(sparse_data_transformed):
     mf = meta_features.helper_functions["Kurtosisses"](
         X_transformed, y, logging.getLogger("Meta")
     )
-    print(mf.value)
     np.testing.assert_allclose(mf.value, fixture)