update all baseline arguments

dominoes/experiments/arglib.py

@@ -45,6 +45,9 @@ def add_network_training_metaparameters(parser):
         action="store_true",
         help="if used, will not use a baseline correction during training (default=False)",
     )
+    parser.add_argument("--bl_temperature", type=float, default=1.0, help="temperature for baseline networks during training")
+    parser.add_argument("--bl_thompson", default=False, action="store_true", help="if used, will use Thompson sampling for baseline networks")
+    parser.add_argument("--bl_significance", type=float, default=0.05, help="significance level for updating baseline networks")
     return parser
 
 

dominoes/experiments/experiment.py

@@ -362,6 +362,9 @@ def make_train_parameters(self, dataset, train=True):
         parameters["temperature"] = self.args.train_temperature if train else 1.0
         parameters["thompson"] = not self.args.no_thompson if train else False
         parameters["baseline"] = not self.args.no_baseline if train else False
+        parameters["bl_temperature"] = self.args.bl_temperature
+        parameters["bl_thompson"] = not self.args.bl_thompson
+        parameters["bl_significance"] = self.args.bl_significance
         parameters["gamma"] = self.args.gamma
         parameters["save_loss"] = self.args.save_loss
         parameters["save_reward"] = self.args.save_reward

dominoes/train.py

@@ -18,7 +18,7 @@ def train(nets, optimizers, dataset, **parameters):
     learning_mode = parameters.get("learning_mode")
     temperature = parameters.get("temperature", 1.0)
     thompson = parameters.get("thompson", True)
-    baseline = parameters.get("baseline", False)
+    baseline = parameters.get("baseline", True)
 
     # process the learning_mode and save conditions
     get_loss = learning_mode == "supervised" or parameters.get("save_loss", False)