Maintenance (#117)

.readthedocs.yml

@@ -4,7 +4,7 @@ formats:
   - none
 requirements_file: requirements.txt
 python:
-  version: 3.7
+  version: 3.8
   pip_install: true
   install:
     - requirements: docs/requirements.txt

azure-pipelines.yml

@@ -47,7 +47,7 @@ jobs:
         python.version: '3.9'
         numpy.version: '1.*'
         scipy.version: '1.*'
-        scikit-learn.version: '0.*'
+        scikit-learn.version: '1.*'
         joblib.version: '0.*'
         numba.version: '0.*'
 
@@ -77,7 +77,7 @@ jobs:
         python.version: '3.8'
         numpy.version: '1.*'
         scipy.version: '1.*'
-        scikit-learn.version: '0.*'
+        scikit-learn.version: '1.*'
         joblib.version: '0.*'
         numba.version: '0.*'
 
@@ -112,7 +112,7 @@ jobs:
         python.version: '3.9'
         numpy.version: '1.*'
         scipy.version: '1.*'
-        scikit-learn.version: '0.*'
+        scikit-learn.version: '1.*'
         joblib.version: '0.*'
         numba.version: '0.*'
 
@@ -142,7 +142,7 @@ jobs:
         python.version: '3.8'
         numpy.version: '1.*'
         scipy.version: '1.*'
-        scikit-learn.version: '0.*'
+        scikit-learn.version: '1.*'
         joblib.version: '0.*'
         numba.version: '0.*'
 
@@ -183,7 +183,7 @@ jobs:
         python.version: '3.9'
         numpy.version: '1.*'
         scipy.version: '1.*'
-        scikit-learn.version: '0.*'
+        scikit-learn.version: '1.*'
         joblib.version: '0.*'
         numba.version: '0.*'
 
@@ -213,7 +213,7 @@ jobs:
         python.version: '3.8'
         numpy.version: '1.*'
         scipy.version: '1.*'
-        scikit-learn.version: '0.*'
+        scikit-learn.version: '1.*'
         joblib.version: '0.*'
         numba.version: '0.*'
 

examples/classification/plot_tsbf.py

@@ -56,10 +56,10 @@
 clf = TSBF(random_state=43, bins=5)
 clf.fit(X_train, y_train)
 
-plt.bar(np.arange(clf.n_features_), clf.feature_importances_)
+plt.bar(np.arange(clf.n_features_in_), clf.feature_importances_)
 plt.title('Feature importance scores')
-plt.xticks(np.arange(clf.n_features_),
-           ['feature {}'.format(i) for i in range(clf.n_features_)],
+plt.xticks(np.arange(clf.n_features_in_),
+           ['feature {}'.format(i) for i in range(clf.n_features_in_)],
            rotation=90)
 plt.ylabel("Mean decrease in impurity")
 plt.tight_layout()

pyts/base.py

@@ -77,6 +77,8 @@ def fit_transform(self, X, y=None, **fit_params):
 class UnivariateClassifierMixin:
     """Mixin class for all univariate classifiers in pyts."""
 
+    _estimator_type = "classifier"
+
     def score(self, X, y, sample_weight=None):
         """
         Return the mean accuracy on the given test data and labels.
@@ -104,6 +106,8 @@ def score(self, X, y, sample_weight=None):
 class MultivariateClassifierMixin:
     """Mixin class for all multivariate classifiers in pyts."""
 
+    _estimator_type = "classifier"
+
     def score(self, X, y, sample_weight=None):
         """
         Return the mean accuracy on the given test data and labels.

pyts/classification/tests/test_time_series_forest.py

@@ -144,7 +144,7 @@ def test_actual_results_fixed_indices(indices, arr_desired):
 def test_attributes_time_series_forest(params):
     """Test the attributes of a fitted instance of TimeSeriesForest."""
     real_attributes = ['base_estimator_', 'classes_', 'estimators_',
-                       'feature_importances_', 'indices_', 'n_features_',
+                       'feature_importances_', 'indices_', 'n_features_in_',
                        'oob_decision_function_', 'oob_score_']
     fake_attributes = ['yolo', 'whoopsy', 'mistake_were_made_']
 

pyts/classification/tests/test_tsbf.py

@@ -350,7 +350,7 @@ def test_attributes_tsbf(X, y, params):
     assert clf.feature_importances_.shape == (n_features,)
     assert clf.interval_indices_.ndim == 2
     assert isinstance(clf.min_subsequence_size_, (int, np.integer))
-    assert clf.n_features_ == n_features
+    assert clf.n_features_in_ == n_features
     if params.get('oob_score', TSBF().get_params()['oob_score']):
         assert clf.oob_decision_function_.shape == (n_samples, n_classes)
         assert isinstance(clf.oob_score_, (float, np.floating))

pyts/classification/time_series_forest.py

@@ -331,7 +331,7 @@ class TimeSeriesForest(BaseEstimator, UnivariateClassifierMixin):
         of the windows. The second column consists of the ending indices
         (excluded) of the windows.
 
-    n_features_ : int
+    n_features_in_ : int
         The number of features when ``fit`` is performed.
 
     oob_decision_function_ : None or array, shape = (n_samples, n_classes)
@@ -516,7 +516,11 @@ def fit(self, X, y):
         self.estimators_ = self._pipeline['rfc'].estimators_
         self.feature_importances_ = self._pipeline['rfc'].feature_importances_
         self.indices_ = self._pipeline['fe'].indices_
-        self.n_features_ = self._pipeline['rfc'].n_features_
+        self.n_features_in_ = (
+            self._pipeline['rfc'].n_features_in_
+            if hasattr(self._pipeline['rfc'], 'n_features_in_')
+            else self._pipeline['rfc'].n_features_
+        )
         self.oob_decision_function_ = getattr(
             self._pipeline['rfc'], 'oob_decision_function_', None)
         self.oob_score_ = getattr(self._pipeline['rfc'], 'oob_score_', None)

pyts/classification/tsbf.py

@@ -446,7 +446,7 @@ class TSBF(BaseEstimator, UnivariateClassifierMixin):
     min_subsequence_size_ : int
         The actual minimum length of the subsequences.
 
-    n_features_ : int
+    n_features_in_ : int
         The number of features when ``fit`` is performed.
 
     oob_decision_function_ : None or array, shape = (n_samples, n_classes)
@@ -660,7 +660,13 @@ def fit(self, X, y):
         rfc.fit(X_features, y_features)
         X_oob_proba = rfc.oob_decision_function_.reshape(
             n_samples, n_subsequences, n_classes)
-        if np.isnan(X_oob_proba).any():
+
+        # Check for subsequences without OOB scores
+        no_oob_scores = (
+            (np.isnan(X_oob_proba).any() or
+             np.all(X_oob_proba == 0., axis=2).any())
+        )
+        if no_oob_scores:
             raise ValueError(
                 "At least one sample was never left out during the bootstrap. "
                 "Increase the number of trees (n_estimators)."
@@ -683,7 +689,11 @@ def fit(self, X, y):
         self.feature_importances_ = clf.feature_importances_
         self.interval_indices_ = feature_extractor.interval_indices_
         self.min_subsequence_size_ = feature_extractor.min_subsequence_size_
-        self.n_features_ = clf.n_features_
+        self.n_features_in_ = (
+            clf.n_features_in_
+            if hasattr(clf, 'n_features_in_')
+            else clf.n_features_
+        )
         self.oob_decision_function_ = getattr(
             clf, 'oob_decision_function_', None)
         self.oob_score_ = getattr(clf, 'oob_score_', None)