FIX_123 (#133)

* FIX_123

* Better debug msg

* at least 1 config in regression

* Return self in _fit()

Author: franchuterivera

autoPyTorch/api/base_task.py

@@ -227,7 +227,7 @@ def set_pipeline_config(
         Args:
             **pipeline_config_kwargs: Valid config options include "num_run",
             "device", "budget_type", "epochs", "runtime", "torch_num_threads",
-            "early_stopping", "use_tensorboard_logger", "use_pynisher",
+            "early_stopping", "use_tensorboard_logger",
             "metrics_during_training"
 
         Returns:

autoPyTorch/configs/default_pipeline_options.json

@@ -7,6 +7,5 @@
             "torch_num_threads": 1,
             "early_stopping": 20,
             "use_tensorboard_logger": "False",
-            "use_pynisher": "False",
             "metrics_during_training": "True"
 }

autoPyTorch/pipeline/base_pipeline.py

@@ -546,6 +546,5 @@ def get_default_pipeline_options() -> Dict[str, Any]:
             'torch_num_threads': 1,
             'early_stopping': 10,
             'use_tensorboard_logger': True,
-            'use_pynisher': False,
             'metrics_during_training': True
         }

autoPyTorch/pipeline/components/training/trainer/base_trainer.py

@@ -29,9 +29,6 @@ def __init__(self,
 
         It also allows to define a 'epoch_or_time' budget type, which means,
         the first of them both which is exhausted, is honored
-
-        In case use_pynisher is set to false, this function allows to
-        still terminate the task with a time domain consideration
         """
         self.start_time = time.time()
         self.budget_type = budget_type

autoPyTorch/pipeline/components/training/trainer/base_trainer_choice.py

@@ -12,8 +12,6 @@
 
 import numpy as np
 
-import pynisher
-
 import torch
 from torch.optim import Optimizer
 from torch.optim.lr_scheduler import _LRScheduler
@@ -196,37 +194,16 @@ def fit(self, X: Dict[str, Any], y: Any = None, **kwargs: Any) -> autoPyTorchCom
                    ] if 'logger_port' in X else logging.handlers.DEFAULT_TCP_LOGGING_PORT,
         )
 
-        fit_function = self._fit
-        if X['use_pynisher']:
-            wall_time_in_s = X['runtime'] if 'runtime' in X else None
-            memory_limit = X['cpu_memory_limit'] if 'cpu_memory_limit' in X else None
-            fit_function = pynisher.enforce_limits(
-                wall_time_in_s=wall_time_in_s,
-                mem_in_mb=memory_limit,
-                logger=self.logger
-            )(self._fit)
-
         # Call the actual fit function.
-        state_dict = fit_function(
+        self._fit(
             X=X,
             y=y,
             **kwargs
         )
 
-        if X['use_pynisher']:
-            # Normally the X[network] is a pointer to the object, so at the
-            # end, when we train using X, the pipeline network is updated for free
-            # If we do multiprocessing (because of pynisher) we have to update
-            # X[network] manually. we do so in a way that every pipeline component
-            # can see this new network -- via an update, not overwrite of the pointer
-            state_dict = state_dict.result
-            X['network'].load_state_dict(state_dict)
-
-        # TODO: when have the optimizer code, the pynisher object might have failed
-        # We should process this function as Failure if so trough fit_function.exit_status
         return cast(autoPyTorchComponent, self.choice)
 
-    def _fit(self, X: Dict[str, Any], y: Any = None, **kwargs: Any) -> torch.nn.Module:
+    def _fit(self, X: Dict[str, Any], y: Any = None, **kwargs: Any) -> 'TrainerChoice':
         """
         Fits a component by using an input dictionary with pre-requisites
 
@@ -359,7 +336,7 @@ def _fit(self, X: Dict[str, Any], y: Any = None, **kwargs: Any) -> torch.nn.Modu
         # Tag as fitted
         self.fitted_ = True
 
-        return X['network'].state_dict()
+        return self
 
     def early_stop_handler(self, X: Dict[str, Any]) -> bool:
         """
@@ -444,10 +421,6 @@ def check_requirements(self, X: Dict[str, Any], y: Any = None) -> None:
             raise ValueError('Need a backend to provide the working directory, '
                              "yet 'backend' was not found in the fit dictionary")
 
-        # For resource allocation, we need to know if pynisher is enabled
-        if 'use_pynisher' not in X:
-            raise ValueError('To fit a Trainer, expected fit dictionary to have use_pynisher')
-
         # Whether we should evaluate metrics during training or no
         if 'metrics_during_training' not in X:
             raise ValueError('Missing metrics_during_training in the fit dictionary')

test/conftest.py

@@ -290,7 +290,6 @@ def get_fit_dictionary(X, y, validator, backend):
         'early_stopping': 10,
         'working_dir': '/tmp',
         'use_tensorboard_logger': True,
-        'use_pynisher': False,
         'metrics_during_training': True,
         'split_id': 0,
         'backend': backend,

test/test_api/test_api.py

@@ -268,7 +268,7 @@ def test_tabular_regression(openml_name, resampling_strategy, backend):
     # Check that smac was able to find proper models
     succesful_runs = [run_value.status for run_value in estimator.run_history.data.values(
     ) if 'SUCCESS' in str(run_value.status)]
-    assert len(succesful_runs) > 1, [(k, v) for k, v in estimator.run_history.data.items()]
+    assert len(succesful_runs) >= 1, [(k, v) for k, v in estimator.run_history.data.items()]
 
     # Search for an existing run key in disc. A individual model might have
     # a timeout and hence was not written to disc

test/test_pipeline/test_tabular_classification.py

@@ -63,7 +63,10 @@ def test_pipeline_fit(self, fit_dictionary_tabular):
         cs = pipeline.get_hyperparameter_search_space()
         config = cs.sample_configuration()
         pipeline.set_hyperparameters(config)
-        pipeline.fit(fit_dictionary_tabular)
+        try:
+            pipeline.fit(fit_dictionary_tabular)
+        except Exception as e:
+            pytest.fail(f"Failed due to {e} for config={config}")
 
         # To make sure we fitted the model, there should be a
         # run summary object with accuracy
@@ -201,7 +204,7 @@ def test_remove_key_check_requirements(self, fit_dictionary_tabular):
         """Makes sure that when a key is removed from X, correct error is outputted"""
         pipeline = TabularClassificationPipeline(
             dataset_properties=fit_dictionary_tabular['dataset_properties'])
-        for key in ['num_run', 'device', 'split_id', 'use_pynisher', 'torch_num_threads', 'dataset_properties']:
+        for key in ['num_run', 'device', 'split_id', 'torch_num_threads', 'dataset_properties']:
             fit_dictionary_tabular_copy = fit_dictionary_tabular.copy()
             fit_dictionary_tabular_copy.pop(key)
             with pytest.raises(ValueError, match=r"To fit .+?, expected fit dictionary to have"):

test/test_pipeline/test_tabular_regression.py

@@ -188,7 +188,7 @@ def test_remove_key_check_requirements(self, fit_dictionary_tabular):
         """Makes sure that when a key is removed from X, correct error is outputted"""
         pipeline = TabularRegressionPipeline(
             dataset_properties=fit_dictionary_tabular['dataset_properties'])
-        for key in ['num_run', 'device', 'split_id', 'use_pynisher', 'torch_num_threads', 'dataset_properties']:
+        for key in ['num_run', 'device', 'split_id', 'torch_num_threads', 'dataset_properties']:
             fit_dictionary_tabular_copy = fit_dictionary_tabular.copy()
             fit_dictionary_tabular_copy.pop(key)
             with pytest.raises(ValueError, match=r"To fit .+?, expected fit dictionary to have"):