Update regression-example-ames-no-preproc.py

Added best parameters yet from hyperparameter optimization study.

Author: david-thrower

regression-example-ames-no-preproc.py

@@ -56,18 +56,18 @@
 # discovered in a bayesian tuning study done on Katib)
 
 meta_trial_number = 0  # In distributed training set this to a random number
-activation = "gelu"
-predecessor_level_connection_affinity_factor_first = 19.613
-predecessor_level_connection_affinity_factor_main = 0.5518
-max_consecutive_lateral_connections = 34
-p_lateral_connection = 0.36014
-num_lateral_connection_tries_per_unit = 11
-learning_rate = 0.095
-epochs = 145
-batch_size = 634
-maximum_levels = 5
-maximum_units_per_level = 5
-maximum_neurons_per_unit = 25
+activation = 'swish'
+predecessor_level_connection_affinity_factor_first = 0.506486683067576
+predecessor_level_connection_affinity_factor_main = 1.6466748663373876
+max_consecutive_lateral_connections = 35
+p_lateral_connection = 3.703218275217572
+num_lateral_connection_tries_per_unit = 12
+learning_rate = 0.02804912925494706
+epochs = 130
+batch_size = 78
+maximum_levels = 4
+maximum_units_per_level = 3
+maximum_neurons_per_unit = 3
 
 
 cerebros =\
@@ -80,11 +80,11 @@
         validation_split=0.35,
         direction='minimize',
         metric_to_rank_by='val_root_mean_squared_error',
-        minimum_levels=1,
+        minimum_levels=4,
         maximum_levels=maximum_levels,
-        minimum_units_per_level=1,
+        minimum_units_per_level=2,
         maximum_units_per_level=maximum_units_per_level,
-        minimum_neurons_per_unit=1,
+        minimum_neurons_per_unit=3,
         maximum_neurons_per_unit=maximum_neurons_per_unit,
         activation=activation,
         final_activation=None,