[temporal] [cont] Fix errors

[test] Add the tests for the instantiation of abstract evaluator 1 -- 3
[test] Add the tests for util 1 -- 2
[test] Add the tests for train_evaluator 1 -- 2
[refactor] [test] Clean up the pipeline classes and add tests for it 1 -- 2
[test] Add the tests for tae 1 -- 4
[fix] Fix an error due to the change in extract learning curve
[experimental] Increase the coverage

[test] Add tests for pipeline repr

Since the modifications in tests removed the coverage on pipeline repr,
I added tests to increase those parts.
Basically, the decrease in the coverage happened due to the usage of
dummy pipelines.

autoPyTorch/api/base_task.py

@@ -1044,7 +1044,7 @@ def _search(
             DisableFileOutputParameters.y_opt in self._disable_file_output
             and self.ensemble_size > 1
         ):
-            self._logger.warning(f"No ensemble will be created when {DisableFileOutputParameters.y_optimization}"
+            self._logger.warning(f"No ensemble will be created when {DisableFileOutputParameters.y_opt}"
                                  f" is in disable_file_output")
 
         self._memory_limit = memory_limit

autoPyTorch/evaluation/abstract_evaluator.py

@@ -193,10 +193,16 @@ def with_default_pipeline_config(
                 f'{cls.__name__}.with_default_pipeline_config() got multiple values for argument `budget_type`'
             )
 
+        budget_type_choices = ('epochs', 'runtime')
         if pipeline_config is None:
             pipeline_config = get_default_pipeline_config(choice=choice)
+        if 'budget_type' not in pipeline_config:
+            raise ValueError('pipeline_config must have `budget_type`')
 
         budget_type = pipeline_config['budget_type']
+        if pipeline_config['budget_type'] not in budget_type_choices:
+            raise ValueError(f"budget_type must be in {budget_type_choices}, but got {budget_type}")
+
         kwargs.update(pipeline_config=pipeline_config, budget_type=budget_type)
         return cls(**kwargs)
 
@@ -307,6 +313,9 @@ def _init_dataset_properties(self) -> None:
                                      ))
 
         self.X_train, self.y_train = datamanager.train_tensors
+        self.unique_train_labels = [
+            list(np.unique(self.y_train[train_indices])) for train_indices, _ in self.splits
+        ]
         self.X_valid, self.y_valid, self.X_test, self.y_test = None, None, None, None
         if datamanager.val_tensors is not None:
             self.X_valid, self.y_valid = datamanager.val_tensors
@@ -377,7 +386,7 @@ def predict(
         self,
         X: Optional[np.ndarray],
         pipeline: BaseEstimator,
-        label_examples: Optional[np.ndarray] = None
+        unique_train_labels: Optional[List[int]] = None
     ) -> Optional[np.ndarray]:
         """
         A wrapper function to handle the prediction of regression or classification tasks.
@@ -387,7 +396,8 @@ def predict(
                 A set of features to feed to the pipeline
             pipeline (BaseEstimator):
                 A model that will take the features X return a prediction y
-            label_examples (Optional[np.ndarray]):
+            unique_train_labels (Optional[List[int]]):
+                The unique labels included in the train split.
 
         Returns:
             (np.ndarray):
@@ -411,7 +421,7 @@ def predict(
                     prediction=pred,
                     num_classes=self.num_classes,
                     output_type=self.output_type,
-                    label_examples=label_examples
+                    unique_train_labels=unique_train_labels
                 )
 
         return pred
@@ -435,6 +445,10 @@ def _get_pipeline(self) -> BaseEstimator:
                 A scikit-learn compliant pipeline which is not yet fit to the data.
         """
         config = self.evaluator_params.configuration
+        if not isinstance(config, (int, str, Configuration)):
+            raise TypeError("The type of configuration must be either (int, str, Configuration), "
+                            f"but got type {type(config)}")
+
         kwargs = dict(
             config=config,
             random_state=np.random.RandomState(self.fixed_pipeline_params.seed),
@@ -452,9 +466,6 @@ def _get_pipeline(self) -> BaseEstimator:
                                   exclude=self.fixed_pipeline_params.exclude,
                                   search_space_updates=self.fixed_pipeline_params.search_space_updates,
                                   **kwargs)
-        else:
-            raise ValueError("The type of configuration must be either (int, str, Configuration), "
-                             f"but got type {type(config)}")
 
     def _loss(self, labels: np.ndarray, preds: np.ndarray) -> Dict[str, float]:
         """SMAC follows a minimization goal, so the make_scorer