Merge pull request #290 from alexbrillant/fix-new-automl-and-choose-o…

…ne-step-of-hp-space

Fix new automl printing, and choose one step of hyperparams spaces

examples/auto_ml_checkpoint.py

@@ -68,7 +68,6 @@ def main(tmpdir, sleep_time: float = 0, n_iter: int = 10):
         pipeline,
         refit_trial=True,
         n_trials=n_iter,
-        print_metrics=False,
         cache_folder_when_no_handle=str(tmpdir),
         validation_split_function=validation_splitter(0.2),
         hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
@@ -106,7 +105,6 @@ def main(tmpdir, sleep_time: float = 0, n_iter: int = 10):
         pipeline,
         refit_trial=False,
         n_trials=n_iter,
-        print_metrics=False,
         cache_folder_when_no_handle=str(tmpdir),
         validation_split_function=validation_splitter(0.2),
         hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),

examples/boston_housing_meta_optimization.py

@@ -97,7 +97,6 @@ def main(tmpdir: LocalPath):
         refit_trial=True,
         n_trials=10,
         epochs=10,
-        print_metrics=False,
         cache_folder_when_no_handle=str(tmpdir),
         scoring_callback=ScoringCallback(mean_squared_error, higher_score_is_better=False),
         callbacks=[MetricCallback('mse', metric_function=mean_squared_error, higher_score_is_better=False)],

neuraxle/__init__.py

@@ -1 +1 @@
-__version__ = "0.3.3"
+__version__ = "0.3.4"

neuraxle/metaopt/auto_ml.py

@@ -464,7 +464,7 @@ def fit(self, p, train_data_container: DataContainer, validation_data_container:
         early_stopping = False
 
         for i in range(self.epochs):
-            self.print_func('epoch {}/{}'.format(i, self.epochs))
+            self.print_func('\nepoch {}/{}'.format(i + 1, self.epochs))
             p = p.handle_fit(train_data_container, context)
 
             y_pred_train = p.handle_predict(train_data_container, context)
@@ -570,7 +570,6 @@ def __init__(
             callbacks: List[BaseCallback] = None,
             refit_scoring_function: Callable = None,
             print_func: Callable = None,
-            print_metrics=True,
             cache_folder_when_no_handle=None
     ):
         BaseStep.__init__(self)
@@ -579,7 +578,6 @@ def __init__(
         self.scoring_callback = scoring_callback
         self.validation_split_function = create_split_data_container_function(validation_split_function)
 
-        self.print_metrics = print_metrics
         if print_func is None:
             print_func = print
 
@@ -631,7 +629,7 @@ def _fit_data_container(self, data_container: DataContainer, context: ExecutionC
         training_data_container, validation_data_container = self.validation_split_function(data_container)
 
         for trial_number in range(self.n_trial):
-            self.print_func('trial {}/{}'.format(trial_number, self.n_trial))
+            self.print_func('\ntrial {}/{}'.format(trial_number + 1, self.n_trial))
 
             auto_ml_data = self._load_auto_ml_data(trial_number)
             p = copy.deepcopy(self.pipeline)
@@ -650,6 +648,7 @@ def _fit_data_container(self, data_container: DataContainer, context: ExecutionC
                     )
 
                     repo_trial.set_success()
+                    self.print_func('trial {}/{} score: {}'.format(trial_number + 1, self.n_trial, repo_trial.get_validation_scores()[-1]))
                 except Exception as error:
                     track = traceback.format_exc()
                     self.print_func(track)
@@ -658,6 +657,8 @@ def _fit_data_container(self, data_container: DataContainer, context: ExecutionC
             self.hyperparams_repository.save_trial(repo_trial)
 
         best_hyperparams = self.hyperparams_repository.get_best_hyperparams()
+
+        self.print_func('best hyperparams:\n{}'.format(json.dumps(best_hyperparams.to_nested_dict(), sort_keys=True, indent=4)))
         p: BaseStep = self._load_virgin_model(hyperparams=best_hyperparams)
         if self.refit_trial:
             p = self.trainer.refit(

neuraxle/metaopt/callbacks.py

@@ -311,10 +311,14 @@ class MetricCallback(BaseCallback):
         :class:`~neuraxle.base.HyperparameterSamples`,
         :class:`~neuraxle.data_container.DataContainer`
     """
-    def __init__(self, name: str, metric_function: Callable, higher_score_is_better: bool):
+    def __init__(self, name: str, metric_function: Callable, higher_score_is_better: bool, print_metrics=True, print_function=None):
         self.name = name
         self.metric_function = metric_function
         self.higher_score_is_better = higher_score_is_better
+        self.print_metrics = print_metrics
+        if print_function is None:
+            print_function = print
+        self.print_function = print_function
 
     def call(self, trial: Trial, epoch_number: int, total_epochs: int, input_train: DataContainer,
              pred_train: DataContainer, input_val: DataContainer, pred_val: DataContainer, is_finished_and_fitted: bool):
@@ -333,6 +337,10 @@ def call(self, trial: Trial, epoch_number: int, total_epochs: int, input_train:
             higher_score_is_better=self.higher_score_is_better
         )
 
+        if self.print_metrics:
+            self.print_function('{} train: {}'.format(self.name, train_score))
+            self.print_function('{} validation: {}'.format(self.name, validation_score))
+
         return False