Fixed issue with hyperparameterm optimozation

Author: RandomDefaultUser

examples/advanced/ex06_distributed_hyperparameter_optimization.py

@@ -36,15 +36,20 @@
 parameters.hyperparameters.study_name = "ex06"
 parameters.hyperparameters.rdb_storage = "sqlite:///ex06.db"
 
-# Hyperparameter optimization can be further refined by using ensemble training
-# at each step and by using a different metric then the validation loss
-# (e.g. the band energy). It is recommended not to use the ensemble training
-# method in Single-GPU use, as it naturally drastically decreases performance.
 parameters.targets.ldos_gridsize = 11
 parameters.targets.ldos_gridspacing_ev = 2.5
 parameters.targets.ldos_gridoffset_ev = -5
 parameters.hyperparameters.number_training_per_trial = 3
-parameters.running.after_training_metric = "band_energy"
+
+# Hyperparameter optimization can be further refined by using ensemble training
+# at each step and by using a different metric then the validation loss
+# (e.g. the band energy). It is recommended not to use the ensemble training
+# method in Single-GPU use, as it naturally drastically decreases performance.
+# For this small example setting, using the band energy as the after training
+# metric is not recommended, since the small data size makes
+# an accurate hyperparameter search difficult. For larger systems, enabling
+# this option is recommended.
+# parameters.running.after_training_metric = "band_energy"
 
 data_handler = mala.DataHandler(parameters)
 

mala/network/trainer.py

@@ -629,7 +629,7 @@ def _validate_network(self, data_set_fractions, metrics):
             The data set fractions to validate on. Can be "train" or
             "validation". "test" is not supported.
 
-        metrics : list
+        metrics : List
             List of metrics to calculate. Can be "ldos", or a number of
             error metrics (see Tester class). Most common apart from "ldos"
             is arguably "band_energy".
@@ -715,10 +715,7 @@ def _validate_network(self, data_set_fractions, metrics):
                 # case for, e.g., distributed network trainings), we can
                 # use a faster (or at least better parallelizing) code
 
-                if (
-                    len(self.parameters.validation_metrics) == 1
-                    and self.parameters.validation_metrics[0] == "ldos"
-                ):
+                if len(metrics) == 1 and metrics[0] == "ldos":
 
                     errors[data_set_type]["ldos"] = (
                         self.__calculate_validation_error_ldos_only(