Bugfix for detecting change points based on MOA (#1613)

* Bugfix based on MOA source code in the _detect_change method. The k index should increment from 0 to bucket.current_idx - 1. The previous code "for k in range(bucket.current_idx - 1):" only increment k to bucket.current_idx - 2 because of the range function.

* Bugfix based on MOA source code in the _detect_change method. The k index should increment from 0 to bucket.current_idx - 1. The previous code "for k in range(bucket.current_idx - 1):" only increment k to bucket.current_idx - 2 because of the range function.

* Tests updated due to bugfix on the ADWIN change detector - reported on issue #1614

river/drift/adwin_c.pyx

@@ -241,7 +241,7 @@ cdef class AdaptiveWindowing:
                         break
                     bucket = self.bucket_deque[idx]
 
-                    for k in range(bucket.current_idx - 1):
+                    for k in range(bucket.current_idx):
                         n2 = self._calculate_bucket_size(idx)   # length of window 2
                         u2 = bucket.get_total_at(k)             # total of window 2
                         # Warning: means are calculated inside the loop to get updated values.
@@ -307,6 +307,7 @@ cdef class AdaptiveWindowing:
                    + (1.0 / (n1 - self.min_window_length + 1)))
         epsilon = (sqrt(2 * m_recip * self.variance_in_window * delta_prime)
                    + 2 / 3 * delta_prime * m_recip)
+
         return fabs(delta_mean) > epsilon
 
 

river/drift/test_drift_detectors.py

@@ -28,7 +28,7 @@
 
 
 def test_adwin():
-    expected_indices = [1055]
+    expected_indices = [1023]
     detected_indices = perform_test(drift.ADWIN(), data_stream_1)
 
     assert detected_indices == expected_indices

river/ensemble/streaming_random_patches.py

@@ -407,7 +407,7 @@ class SRPClassifier(BaseSRPEnsemble, base.Classifier):
     >>> metric = metrics.Accuracy()
 
     >>> evaluate.progressive_val_score(dataset, model, metric)
-    Accuracy: 71.97%
+    Accuracy: 72.17%
 
     Notes
     -----

river/forest/adaptive_random_forest.py

@@ -565,7 +565,7 @@ class ARFClassifier(BaseForest, base.Classifier):
     >>> metric = metrics.Accuracy()
 
     >>> evaluate.progressive_val_score(dataset, model, metric)
-    Accuracy: 71.17%
+    Accuracy: 67.97%
 
     The total number of warnings and drifts detected, respectively
     >>> model.n_warnings_detected(), model.n_drifts_detected()
@@ -849,7 +849,7 @@ class ARFRegressor(BaseForest, base.Regressor):
     >>> metric = metrics.MAE()
 
     >>> evaluate.progressive_val_score(dataset, model, metric)
-    MAE: 0.788619
+    MAE: 0.772113
 
     """
 

river/forest/online_extra_trees.py

@@ -614,7 +614,7 @@ class OXTRegressor(ExtraTrees, base.Regressor):
     >>> metric = metrics.RMSE()
 
     >>> evaluate.progressive_val_score(dataset, model, metric)
-    RMSE: 3.127311
+    RMSE: 3.16212
 
     References
     ----------

river/imblearn/chebyshev.py

@@ -162,10 +162,10 @@ class ChebyshevOverSampler(base.Wrapper, base.Regressor):
     ...     metrics.MAE(),
     ...     print_every=500
     ... )
-    [500] MAE: 1.673902
-    [1,000] MAE: 1.743046
-    [1,001] MAE: 1.741335
-    MAE: 1.741335
+    [500] MAE: 1.629786
+    [1,000] MAE: 1.663799
+    [1,001] MAE: 1.66253
+    MAE: 1.66253
 
     References
     ----------

river/tree/hoeffding_adaptive_tree_regressor.py

@@ -140,7 +140,7 @@ class HoeffdingAdaptiveTreeRegressor(HoeffdingTreeRegressor):
     >>> metric = metrics.MAE()
 
     >>> evaluate.progressive_val_score(dataset, model, metric)
-    MAE: 0.823026
+    MAE: 0.917576
 
     """