Automatic per run time limit selection (#884)

* Automatic per run time limit selection

* Spell-fix

* Incorporating comments from commit

* Flake8

* Remove slack for ensemble builder (Not needed)

Author: franchuterivera

autosklearn/automl.py

@@ -433,13 +433,26 @@ def fit(
         else:
             if self._per_run_time_limit is None or \
                     self._per_run_time_limit > time_left_for_smac:
-                print('Time limit for a single run is higher than total time '
-                      'limit. Capping the limit for a single run to the total '
-                      'time given to SMAC (%f)' % time_left_for_smac)
+                self._logger.warning(
+                    'Time limit for a single run is higher than total time '
+                    'limit. Capping the limit for a single run to the total '
+                    'time given to SMAC (%f)' % time_left_for_smac
+                )
                 per_run_time_limit = time_left_for_smac
             else:
                 per_run_time_limit = self._per_run_time_limit
 
+            # Make sure that at least 2 models are created for the ensemble process
+            num_models = time_left_for_smac // per_run_time_limit
+            if num_models < 2:
+                per_run_time_limit = time_left_for_smac//2
+                self._logger.warning(
+                    "Capping the per_run_time_limit to {} to have "
+                    "time for a least 2 models in each process.".format(
+                        per_run_time_limit
+                    )
+                )
+
             _proc_smac = AutoMLSMBO(
                 config_space=self.configuration_space,
                 dataset_name=self._dataset_name,

autosklearn/estimators.py

@@ -21,7 +21,7 @@ class AutoSklearnEstimator(BaseEstimator):
     def __init__(
         self,
         time_left_for_this_task=3600,
-        per_run_time_limit=360,
+        per_run_time_limit=None,
         initial_configurations_via_metalearning=25,
         ensemble_size: int = 50,
         ensemble_nbest=50,
@@ -56,7 +56,7 @@ def __init__(
             models. By increasing this value, *auto-sklearn* has a higher
             chance of finding better models.
 
-        per_run_time_limit : int, optional (default=360)
+        per_run_time_limit : int, optional (default=1/10 of time_left_for_this_task)
             Time limit for a single call to the machine learning model.
             Model fitting will be terminated if the machine learning
             algorithm runs over the time limit. Set this value high enough so
@@ -344,8 +344,19 @@ def fit(self, **kwargs):
                 'only one of them.'
             )
 
+        # Handle the number of jobs and the time for them
         if self.n_jobs is None or self.n_jobs == 1:
             self._n_jobs = 1
+        elif self.n_jobs == -1:
+            self._n_jobs = joblib.cpu_count()
+        else:
+            self._n_jobs = self.n_jobs
+
+        # Automatically set the cutoff time per task
+        if self.per_run_time_limit is None:
+            self.per_run_time_limit = self._n_jobs * self.time_left_for_this_task // 10
+
+        if self.n_jobs is None or self.n_jobs == 1:
             shared_mode = self.shared_mode
             seed = self.seed
             automl = self.build_automl(
@@ -366,11 +377,6 @@ def fit(self, **kwargs):
                 output_directory=self.output_folder,
             )
 
-            if self.n_jobs == -1:
-                self._n_jobs = joblib.cpu_count()
-            else:
-                self._n_jobs = self.n_jobs
-
             shared_mode = True
             seeds = set()
             for i in range(self._n_jobs):