Updates to support Python 3.7

microsoft · Dec 17, 2023 · 2fe1d2f · 2fe1d2f
1 parent b2869dc
commit 2fe1d2f
Show file tree

Hide file tree

Showing 2 changed files with 26 additions and 13 deletions.
diff --git a/raimitigations/automitigator/evaluator.py b/raimitigations/automitigator/evaluator.py
@@ -14,21 +14,20 @@ class Evaluator:
 
     def __init__(self, automl_args=None) -> None:
         self.automl_args = automl_args
+        self.pipeline_steps = []
         self.pipeline = None
 
     def _pipeline_append(self, step):
         """
         Append a step to the pipeline.
         Note: This is done to support older versions of sklearn < 1.1 (required to support python 3.7)
-        which doesn't support an empty pipeline to be initialized.
+        which doesn't support an empty pipeline to be initialized nor appending a step to an existing pipeline
+        without an estimator at the end.
 
         :param sklearn.pipeline.Pipeline pipeline: The pipeline to add the step to
         :param step: The step to add
         """
-        if self.pipeline is None:
-            self.pipeline = Pipeline([step])
-        else:
-            self.pipeline.steps.append(step)
+        self.pipeline_steps.append(step)
 
     def evaluate(self, train_x, train_y, search_config):
         """
@@ -52,8 +51,6 @@ def evaluate(self, train_x, train_y, search_config):
         cohort = search_space[amd.cohort_key]
         if cohort == amd.all_cohort:
             return self.mitigate_full_dataset(train_x, train_y, search_space)
-        else:
-            raise ValueError(f"Unknown cohort type {cohort}")
 
     def _process_feature_selector(self, selector_type):
         """
@@ -131,8 +128,6 @@ def mitigate_full_dataset(self, train_x, train_y, search_space):
                 self._process_feature_selector(config[amd.mitigation_type_key])
             elif mitigation_name == amd.no_mitigation:
                 continue
-            else:
-                raise ValueError(f"Unknown mitigation {mitigation_name}")
 
         fit_results = self._fit_model(train_x, train_y)
         fit_results["search_space"] = search_space
@@ -153,8 +148,6 @@ def _process_imputer(self, imputer_type):
             self._pipeline_append((amd.iterative_imputer, dp.IterativeDataImputer()))
         elif imputer_name == amd.knn_imputer:
             self._pipeline_append((amd.knn_imputer, dp.KNNDataImputer()))
-        else:
-            raise ValueError(f"Unknown imputer {imputer_name}")
 
     def _process_scaler(self, scaler_type):
         """
@@ -177,8 +170,6 @@ def _process_scaler(self, scaler_type):
             self._pipeline_append((amd.normalize_scaler, dp.DataNormalizer()))
         elif scaler_name == amd.minmax_scaler:
             self._pipeline_append((amd.minmax_scaler, dp.DataMinMaxScaler()))
-        else:
-            raise ValueError(f"Unknown scaler {scaler_name}")
 
     def _fit_model(self, train_x, train_y):
         """
@@ -197,6 +188,7 @@ def _fit_model(self, train_x, train_y):
         """
         automl = AutoML(**self.automl_args)
         self._pipeline_append(("automl", automl))
+        self.pipeline = Pipeline(self.pipeline_steps)
 
         try:
             self.pipeline.fit(train_x, train_y)

diff --git a/test/automitigator/automitigator_test.py b/test/automitigator/automitigator_test.py
@@ -44,6 +44,27 @@ def _test_mitigation(train_x, train_y, test_x, mitigation):
 
     assert autoMitigator._pipeline is not None
 
+# -----------------------------------
+def test_duplicate_mitigations(df_breast_cancer, label_name_bc):
+    train_x, test_x, train_y, test_y = _prepare_data(df_breast_cancer, label_name_bc)
+
+    config = {
+        'search_space':
+        {
+            'cohort': 'all',
+            'mitigations': { 'action0': {'type': 0, 'strategy': 0, 'name': 'rebalancer'},
+                             'action1': {'type': 0, 'strategy': 0, 'name': 'rebalancer'}}
+        }
+    }
+
+    tune_args = {'points_to_evaluate':[config]}
+    autoMitigator = AutoMitigator(max_mitigations=2, num_samples=1, use_ray=False, tune_args=tune_args)
+    autoMitigator.fit(train_x, train_y)
+    _ = autoMitigator.predict(test_x)
+
+    assert autoMitigator._pipeline is not None
+
+# -----------------------------------
 def test_auto_mitigator_starting_points(df_breast_cancer, label_name_bc):
     train_x, test_x, train_y, test_y = _prepare_data(df_breast_cancer, label_name_bc)