fixed test for storing callback results in evaluator

src/dbispipeline/evaluators.py

@@ -62,6 +62,7 @@ class ModelCallbackWrapper(Evaluator):
             self.callbacks = callback
         self.callback_args = callback_args
         self.callback_kwargs = callback_kwargs
+        self.callback_results = {}
 
     def evaluate(self, model, data):
         """
@@ -77,16 +78,17 @@ class ModelCallbackWrapper(Evaluator):
         result = self.wrapped_evaluator.evaluate(model, data)
         fitted_model = self.wrapped_evaluator.fitted_model
         for callback in self.callbacks:
-            callback(fitted_model, *self.callback_args, **self.callback_kwargs)
+            self.callback_results[callback.__name__] = callback(
+                fitted_model, *self.callback_args, **self.callback_kwargs)
         return result
 
     @property
     def configuration(self):
         """A json-like representation of the configuration."""
         return {
-            'model_callback': self.callbacks.__name__,
             'wrapped_evaluator': self.wrapped_evaluator.__class__.__name__,
             'wrapped_configuration': self.wrapped_evaluator.configuration,
+            'callbacks': self.callback_results,
         }
 
 

tests/evaluators/test_model_storing.py

@@ -22,7 +22,7 @@ def test_model_store_evaluator():
                                      pickle_model_callback,
                                      output_path=tmp.name)
     evaluator.evaluate(pipeline, data)
-    actual_filename = evaluator.callback_result_
+    actual_filename = evaluator.callback_results['pickle_model_callback']
     with open(actual_filename, 'rb') as i_f:
         model = pickle.load(i_f)
         # the equality of the SVM weights should suffice as valid indicator.