pass dataset instead of dataset info to autonet.get_hyperparameter_search_space()

Author: urbanmatthias

README.md

@@ -95,7 +95,8 @@ autoPyTorch = AutoNetClassification(networks=["resnet", "shapedresnet", "mlpnet"
 # Each hyperparameter belongs to a node in Auto-PyTorch's ML Pipeline.
 # The names of the hyperparameters are prefixed with the name of the node: NodeName:hyperparameter_name.
 # If a hyperparameter belongs to a component: NodeName:component_name:hyperparameter_name.
-autoPyTorch.get_hyperparameter_search_space()
+# Call with the same arguments as fit.
+autoPyTorch.get_hyperparameter_search_space(X_train, y_train, validation_split=0.3)
 
 # You can configure the search space of every hyperparameter of every component:
 from autoPyTorch import HyperparameterSearchSpaceUpdates

autoPyTorch/components/training/trainer.py

@@ -88,7 +88,7 @@ def final_eval(self, opt_metric_name, logs, train_loader, valid_loader, minimize
             for i, metric in enumerate(self.metrics):
                 if valid_metric_results:
                     final_log['val_' + metric.__name__] = valid_metric_results[i]
-            if self.eval_additional_logs_on_snapshot:
+            if self.eval_additional_logs_on_snapshot and not refit:
                     for additional_log in self.log_functions:
                         final_log[additional_log.__name__] = additional_log(self.model, None)
         return final_log

autoPyTorch/core/api.py

@@ -15,6 +15,7 @@
 from autoPyTorch.pipeline.nodes.cross_validation import CrossValidation
 from autoPyTorch.pipeline.nodes.metric_selector import MetricSelector
 from autoPyTorch.pipeline.nodes.optimization_algorithm import OptimizationAlgorithm
+from autoPyTorch.pipeline.nodes.create_dataset_info import CreateDatasetInfo
 
 
 from autoPyTorch.utils.config.config_file_parser import ConfigFileParser
@@ -68,13 +69,28 @@ def get_current_autonet_config(self):
             return self.autonet_config
         return self.pipeline.get_pipeline_config(**self.base_config)
 
-    def get_hyperparameter_search_space(self, dataset_info=None):
-        """Return the hyperparameter search space of AutoNet
+    def get_hyperparameter_search_space(self, X_train=None, Y_train=None, X_valid=None, Y_valid=None, **autonet_config):
+        """Return hyperparameter search space of Auto-PyTorch. Does depend on the dataset!
+        
+        Keyword Arguments:
+            X_train {array} -- Training data.
+            Y_train {array} -- Targets of training data.
+            X_valid {array} -- Validation data. Will be ignored if cv_splits > 1. (default: {None})
+            Y_valid {array} -- Validation data. Will be ignored if cv_splits > 1. (default: {None})
         
         Returns:
-            ConfigurationSpace -- The ConfigurationSpace that should be optimized
+            ConfigurationSpace -- The configuration space that should be optimized.
         """
 
+        dataset_info = None
+        if X_train is not None and Y_train is not None:
+            dataset_info_node = self.pipeline[CreateDatasetInfo.get_name()]
+            dataset_info = dataset_info_node.fit(pipeline_config=dict(self.base_config, **autonet_config),
+                                                 X_train=X_train,
+                                                 Y_train=Y_train,
+                                                 X_valid=X_valid,
+                                                 Y_valid=Y_valid)["dataset_info"]
+
         return self.pipeline.get_hyperparameter_search_space(dataset_info=dataset_info, **self.get_current_autonet_config())
 
     @classmethod