trim smote infinite loop fix

analyticalmindsltd · Aug 22, 2022 · 6e7286b · 6e7286b
1 parent 212b8c2
commit 6e7286b
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 4 deletions.
diff --git a/smote_variants/evaluation/_evaluation.py b/smote_variants/evaluation/_evaluation.py
@@ -147,7 +147,9 @@ def do_evaluation(self):
                     'classifier_params': classifier[2],
                     'classifier_module': classifier[0],
                     'error': error,
-                    'warning': str(warning_list)}
+                    'warning': str(warning_list),
+                    'oversampling_error': oversampling['error'],
+                    'oversampling_warning': oversampling['warning']}
 
             if self.cache_path is not None:
                 dump_dict(evaluation, target_filename, self.serialization,

diff --git a/smote_variants/oversampling/_trim_smote.py b/smote_variants/oversampling/_trim_smote.py
@@ -337,12 +337,9 @@ def generate_samples(self, X, y,
         nnmt.fit(X_seed_min)
         indices = nnmt.kneighbors(X_seed_min, return_distance=False)
 
-        #n_dim_orig = self.n_dim
-        #self.n_dim = np.min([self.n_dim, X_seed_min.shape[0]])
         samples = self.sample_simplex(X=X_seed_min,
                                       indices=indices,
                                       n_to_sample=n_to_sample)
-        #self.n_dim = n_dim_orig
 
         return samples
 
@@ -364,6 +361,9 @@ def sampling_algorithm(self, X, y):
 
         seeds = self.trimming(X, y)
 
+        if len([s for s in seeds if self.precision(s[1]) > self.min_precision/10.0]) == 0:
+            return self.return_copies(X, y, "no seeds found")
+
         # filtering the resulting set
         filtered_seeds = [s for s in seeds if self.precision(s[1]) > self.min_precision]
 

diff --git a/tests/oversampling/algorithm_level/test_trim_smote.py b/tests/oversampling/algorithm_level/test_trim_smote.py
@@ -20,6 +20,12 @@ def test_specific():
 
     assert len(X_samp) > 0
 
+    obj.min_precision = 10000.0
+
+    X_samp, _ = obj.sample(dataset['data'], dataset['target'])
+
+    assert len(X_samp) > 0
+
     assert len(obj.parameter_combinations(raw=True)) > 0
 
     X = np.array([[1, 2], [2, 3], [3, 4], [4, 5],