fix: missing tests of anomaly detectors (#2102)

aeon/anomaly_detection/_dwt_mlead.py

@@ -134,8 +134,8 @@ def _predict(self, X) -> np.ndarray:
         if self.quantile_epsilon < 0 or self.quantile_epsilon > 1:
             raise ValueError("quantile_epsilon must be in [0, 1]")
 
-        X, self.N_, self.M_ = _pad_series(X)
-        max_level = int(np.log2(self.M_))
+        X, n, m = _pad_series(X)
+        max_level = int(np.log2(m))
 
         if self.start_level >= max_level:
             raise ValueError(
@@ -168,7 +168,7 @@ def _predict(self, X) -> np.ndarray:
 
         # aggregate anomaly counts (leaf counters)
         point_anomaly_scores = self._push_anomaly_counts_down_to_points(
-            coef_anomaly_counts
+            coef_anomaly_counts, m, n
         )
         return point_anomaly_scores
 
@@ -220,19 +220,20 @@ def _reverse_windowing(
 
         return np.sum(mapped, axis=1)
 
+    @staticmethod
     def _push_anomaly_counts_down_to_points(
-        self, coef_anomaly_counts: list[np.ndarray]
+        coef_anomaly_counts: list[np.ndarray], m: int, n: int
     ) -> np.ndarray:
         # sum up counters of detail coeffs (orig. D^l) and approx coeffs (orig. C^l)
         anomaly_counts = coef_anomaly_counts[0::2] + coef_anomaly_counts[1::2]
 
         # extrapolate anomaly counts to the original series' points
-        counter = np.zeros(self.M_)
+        counter = np.zeros(m)
         for ac in anomaly_counts:
-            counter += ac.repeat(self.M_ // ac.shape[0], axis=0)
+            counter += ac.repeat(m // ac.shape[0], axis=0)
         # set event counters with count < 2 to 0
         counter[counter < 2] = 0
-        return counter[: self.N_]
+        return counter[:n]
 
     @classmethod
     def get_test_params(cls, parameter_set="default"):

aeon/anomaly_detection/_pyodadapter.py

@@ -8,6 +8,7 @@
 from typing import TYPE_CHECKING, Any
 
 import numpy as np
+from sklearn import clone
 
 from aeon.anomaly_detection.base import BaseAnomalyDetector
 from aeon.utils.validation._dependencies import _check_soft_dependencies
@@ -114,7 +115,7 @@ def _predict(self, X: np.ndarray) -> np.ndarray:
         _X, padding = sliding_windows(
             X, window_size=self.window_size, stride=self.stride, axis=0
         )
-        window_anomaly_scores = self.pyod_model.decision_function(_X)
+        window_anomaly_scores = self.fitted_pyod_model_.decision_function(_X)
         point_anomaly_scores = reverse_windowing(
             window_anomaly_scores, self.window_size, np.nanmean, self.stride, padding
         )
@@ -127,7 +128,7 @@ def _fit_predict(self, X: np.ndarray, y: np.ndarray | None = None) -> np.ndarray
         )
         self._inner_fit(_X)
 
-        window_anomaly_scores = self.pyod_model.decision_scores_
+        window_anomaly_scores = self.fitted_pyod_model_.decision_scores_
         point_anomaly_scores = reverse_windowing(
             window_anomaly_scores, self.window_size, np.nanmean, self.stride, padding
         )
@@ -149,10 +150,11 @@ def _check_params(self, X: np.ndarray) -> None:
             )
 
     def _inner_fit(self, X: np.ndarray) -> None:
-        self.pyod_model.fit(X)
+        self.fitted_pyod_model_: BaseDetector = clone(self.pyod_model)  # type: ignore
+        self.fitted_pyod_model_.fit(X)
 
     def _inner_predict(self, X: np.ndarray, padding: int) -> np.ndarray:
-        window_anomaly_scores = self.pyod_model.decision_function(X)
+        window_anomaly_scores = self.fitted_pyod_model_.decision_function(X)
         point_anomaly_scores = reverse_windowing(
             window_anomaly_scores, self.window_size, np.nanmean, self.stride, padding
         )

aeon/anomaly_detection/_stomp.py

@@ -82,7 +82,6 @@ def __init__(
         p: float = 2.0,
         k: int = 1,
     ):
-        self.mp: np.ndarray | None = None
         self.window_size = window_size
         self.ignore_trivial = ignore_trivial
         self.normalize = normalize
@@ -96,15 +95,15 @@ def _predict(self, X: np.ndarray) -> np.ndarray:
         import stumpy
 
         self._check_params(X)
-        self.mp = stumpy.stump(
+        mp = stumpy.stump(
             X[:, 0],
             m=self.window_size,
             ignore_trivial=self.ignore_trivial,
             normalize=self.normalize,
             p=self.p,
             k=self.k,
         )
-        point_anomaly_scores = reverse_windowing(self.mp[:, 0], self.window_size)
+        point_anomaly_scores = reverse_windowing(mp[:, 0], self.window_size)
         return point_anomaly_scores
 
     def _check_params(self, X: np.ndarray) -> None:

aeon/testing/testing_config.py

@@ -58,11 +58,8 @@
     "RSAST": ["check_fit_deterministic"],
     "AEFCNClusterer": ["check_fit_updates_state"],
     "AEResNetClusterer": ["check_fit_updates_state"],
-    "PyODAdapter": ["check_fit_updates_state"],
     "SFA": ["check_persistence_via_pickle", "check_fit_deterministic"],
     # missed in legacy testing, changes state in predict/transform
-    "DWT_MLEAD": ["check_non_state_changing_method"],
-    "STOMP": ["check_non_state_changing_method"],
     "FLUSSSegmenter": ["check_non_state_changing_method"],
     "InformationGainSegmenter": ["check_non_state_changing_method"],
     "GreedyGaussianSegmenter": ["check_non_state_changing_method"],
@@ -84,9 +81,8 @@
 # methods that should not change the state of the estimator, that is, they should
 # not change fitted parameters or hyper-parameters. They are also the methods that
 # "apply" the fitted estimator to data and useful for checking results.
-# NON_STATE_CHANGING_METHODS_ARRAYLIK =
-# non-state-changing methods that return an array-like output
 
+# non-state-changing methods that return an array-like output
 NON_STATE_CHANGING_METHODS_ARRAYLIKE = (
     "predict",
     "predict_var",