added tests to make sure the trainers receive the right parameters

python/learn.py

@@ -114,6 +114,10 @@ def get_progress():
             if 'is_ghost' not in trainer_parameters_dict[brain_name]:
                 trainer_parameters_dict[brain_name]['is_ghost'] = False
             if trainer_parameters_dict[brain_name]['is_ghost']:
+                if trainer_parameters_dict[brain_name]['brain_to_copy'] not in env.external_brain_names:
+                    raise UnityEnvironmentException("The external brain {0} could not be found in the environment "
+                      "even though the ghost trainer of brain {1} is trying to ghost it."
+                      .format(trainer_parameters_dict[brain_name]['brain_to_copy'], brain_name))
                 trainer_parameters_dict[brain_name]['original_brain_parameters'] = trainer_parameters_dict[
                     trainer_parameters_dict[brain_name]['brain_to_copy']]
                 trainers[brain_name] = GhostTrainer(sess, env, brain_name, trainer_parameters_dict[brain_name], train_model)

python/trainer_configurations.yaml

@@ -31,3 +31,5 @@ ghost-Ball3DBrain:
     is_ghost : true
     new_model_freq : 10000
     max_num_models : 3
+Ball3DBrain:
+    is_imitation : true

python/trainers/ghost_trainer.py

@@ -19,8 +19,13 @@ def __init__(self, sess, env, brain_name, trainer_parameters, training):
         :param  trainer_parameters: The parameters for the trainer (dictionary).
         :param training: Whether the trainer is set for training.
         """
+        self.param_keys = ['brain_to_copy', 'is_ghost', 'new_model_freq', 'max_num_models']
+        for k in self.param_keys:
+            if k not in trainer_parameters:
+                raise UnityEnvironmentException("The hyperparameter {0} could not be found for the PPO trainer of "
+                    "brain {1}.".format(k, brain_name))
+
 
-        # TODO: check validity of parameters
         self.brain_to_copy = trainer_parameters['brain_to_copy']
         self.variable_scope = trainer_parameters['graph_scope']
         self.original_brain_parameters = trainer_parameters['original_brain_parameters']
@@ -59,9 +64,8 @@ def __init__(self, sess, env, brain_name, trainer_parameters, training):
         self.trainer_parameters = trainer_parameters
 
     def __str__(self):
-        param_keys = ['brain_to_copy', 'is_ghost', 'new_model_freq', 'max_num_models']
         return '''Hypermarameters for {0}: \n{1}'''.format(
-            self.brain_name, '\n'.join(['\t{0} :\t{1}'.format(x, self.trainer_parameters[x]) for x in param_keys]))
+            self.brain_name, '\n'.join(['\t{0} :\t{1}'.format(x, self.trainer_parameters[x]) for x in self.param_keys]))
 
 
     @property

python/trainers/ppo_trainer.py

@@ -10,6 +10,7 @@
 
 from trainers.buffer import Buffer
 from trainers.ppo_models import *
+from unityagents import UnityEnvironmentException
 
 logger = logging.getLogger("unityagents")
 
@@ -25,6 +26,15 @@ def __init__(self, sess, env, brain_name, trainer_parameters, training):
         :param  trainer_parameters: The parameters for the trainer (dictionary).
         :param training: Whether the trainer is set for training.
         """
+        self.param_keys = ['batch_size', 'beta','buffer_size','epsilon','gamma','hidden_units','lambd','learning_rate',
+            'max_steps','normalize','num_epoch','num_layers','time_horizon','sequence_length','summary_freq',
+            'use_recurrent','graph_scope','summary_path']
+
+        for k in self.param_keys:
+            if k not in trainer_parameters:
+                raise UnityEnvironmentException("The hyperparameter {0} could not be found for the PPO trainer of "
+                    "brain {1}.".format(k, brain_name))
+
         self.use_recurrent = trainer_parameters["use_recurrent"]
         self.sequence_length = 1
         self.m_size = None
@@ -67,11 +77,8 @@ def __init__(self, sess, env, brain_name, trainer_parameters, training):
         self.trainer_parameters = trainer_parameters
 
     def __str__(self):
-        param_keys = ['batch_size', 'beta','buffer_size','epsilon','gamma','hidden_units','lambd','learning_rate',
-            'max_steps','normalize','num_epoch','num_layers','time_horizon','sequence_length','summary_freq',
-            'use_recurrent','graph_scope','summary_path']
         return '''Hypermarameters for {0}: \n{1}'''.format(
-            self.brain_name, '\n'.join(['\t{0} :\t{1}'.format(x, self.trainer_parameters[x]) for x in param_keys]))
+            self.brain_name, '\n'.join(['\t{0} :\t{1}'.format(x, self.trainer_parameters[x]) for x in self.param_keys]))
 
     @property
     def parameters(self):