[ENH] adjust test for non numpy output (#2517)

* adjust test for non numpy output

* test list output

* test dataframe output

* change pickle test

* equal nans

* test scalar output

* fix lists output

* allow arrays of objects

* allow arrays of objects

* test for boolean elements (MERLIN)

* switch to deep equals

* switch to deep equals

* switch to deep equals

* switch to deep equals

* switch to deep equals

* add np.bool_

* rework hydra classifier for numpy

* rework hydra classifier for numpy

* exclude HydraTransformer

* unexclude SFA

* remove line

aeon/testing/estimator_checking/_yield_estimator_checks.py

@@ -9,7 +9,6 @@
 
 import joblib
 import numpy as np
-from numpy.testing import assert_array_almost_equal
 from sklearn.exceptions import NotFittedError
 
 from aeon.anomaly_detection.base import BaseAnomalyDetector
@@ -625,20 +624,19 @@ def check_persistence_via_pickle(estimator, datatype):
     for method in NON_STATE_CHANGING_METHODS_ARRAYLIKE:
         if hasattr(estimator, method) and callable(getattr(estimator, method)):
             output = _run_estimator_method(estimator, method, datatype, "test")
-            assert_array_almost_equal(
-                output,
-                results[i],
-                err_msg=f"Running {method} after fit twice with test "
-                f"parameters gives different results.",
-            )
+            same, msg = deep_equals(output, results[i], return_msg=True)
+            if not same:
+                raise ValueError(
+                    f"Running {method} after serialisation parameters gives "
+                    f"different results. "
+                    f"{type(estimator)} returns data as {type(output)}: test "
+                    f"equivalence message: {msg}"
+                )
             i += 1
 
 
 def check_fit_deterministic(estimator, datatype):
-    """Test that fit is deterministic.
-
-    Check that calling fit twice is equivalent to calling it once.
-    """
+    """Check that calling fit twice is equivalent to calling it once."""
     estimator = _clone_estimator(estimator, random_state=0)
     _run_estimator_method(estimator, "fit", datatype, "train")
 
@@ -648,17 +646,20 @@ def check_fit_deterministic(estimator, datatype):
             output = _run_estimator_method(estimator, method, datatype, "test")
             results.append(output)
 
-    # run fit and other methods a second time
+    # run fit a second time
     _run_estimator_method(estimator, "fit", datatype, "train")
 
+    # check output of predict/transform etc does not change
     i = 0
     for method in NON_STATE_CHANGING_METHODS_ARRAYLIKE:
         if hasattr(estimator, method) and callable(getattr(estimator, method)):
             output = _run_estimator_method(estimator, method, datatype, "test")
-            assert_array_almost_equal(
-                output,
-                results[i],
-                err_msg=f"Running {method} after fit twice with test "
-                f"parameters gives different results.",
-            )
+            same, msg = deep_equals(output, results[i], return_msg=True)
+            if not same:
+                raise ValueError(
+                    f"Running {method} with test parameters after two calls to fit "
+                    f"gives different results."
+                    f"{type(estimator)} returns data as {type(output)}: test "
+                    f"equivalence message: {msg}"
+                )
             i += 1

aeon/testing/testing_config.py

@@ -23,7 +23,8 @@
 NUMBA_DISABLED = os.environ.get("NUMBA_DISABLE_JIT") == "1"
 
 # exclude estimators here for short term fixes
-EXCLUDE_ESTIMATORS = ["REDCOMETS"]
+# Hydra excluded because it returns a pytorch Tensor
+EXCLUDE_ESTIMATORS = ["REDCOMETS", "HydraTransformer"]
 
 # Exclude specific tests for estimators here
 EXCLUDED_TESTS = {
@@ -49,7 +50,6 @@
     "RSASTClassifier": ["check_fit_deterministic"],
     "SAST": ["check_fit_deterministic"],
     "RSAST": ["check_fit_deterministic"],
-    "SFA": ["check_persistence_via_pickle", "check_fit_deterministic"],
     # missed in legacy testing, changes state in predict/transform
     "FLUSSSegmenter": ["check_non_state_changing_method"],
     "InformationGainSegmenter": ["check_non_state_changing_method"],

aeon/testing/utils/deep_equals.py

@@ -72,7 +72,7 @@ def _deep_equals(x, y, depth, ignore_index):
         eq = np.isnan(y)
         msg = "" if eq else f"x ({x}) != y ({y}), depth={depth}"
         return eq, msg
-    elif isinstance(x == y, bool):
+    elif isinstance(x == y, (bool, np.bool_)):
         eq = x == y
         msg = "" if eq else f"x ({x}) != y ({y}), depth={depth}"
         return eq, msg
@@ -131,9 +131,11 @@ def _dataframe_equals(x, y, depth, ignore_index):
 def _numpy_equals(x, y, depth):
     if x.dtype != y.dtype:
         return False, f"x.dtype ({x.dtype}) != y.dtype ({y.dtype})"
-
-    eq = np.allclose(x, y, equal_nan=True)
-    msg = "" if eq else f"x ({x}) != y ({y}), depth={depth}"
+    if x.dtype == "object":
+        eq, msg = _deep_equals(x.tolist(), y.tolist(), depth, ignore_index=True)
+    else:
+        eq = np.allclose(x, y, equal_nan=True)
+        msg = "" if eq else f"x ({x}) != y ({y}), depth={depth}"
     return eq, msg