[refactor] Remove references to brain_name in policy

docs/Training-Configuration-File.md

@@ -154,7 +154,7 @@ A few considerations when deciding to use memory:
   too large `memory_size` will slow down training.
 - Adding a recurrent layer increases the complexity of the neural network, it is
   recommended to decrease `num_layers` when using recurrent.
-- It is required that `memory_size` be divisible by 4.
+- It is required that `memory_size` be divisible by 2.
 
 ## Self-Play
 

ml-agents/mlagents/trainers/policy/tf_policy.py

@@ -95,18 +95,6 @@ def __init__(
         if self.network_settings.memory is not None:
             self.m_size = self.network_settings.memory.memory_size
             self.sequence_length = self.network_settings.memory.sequence_length
-            if self.m_size == 0:
-                raise UnityPolicyException(
-                    "The memory size for brain {0} is 0 even "
-                    "though the trainer uses recurrent.".format(brain.brain_name)
-                )
-            elif self.m_size % 2 != 0:
-                raise UnityPolicyException(
-                    "The memory size for brain {0} is {1} "
-                    "but it must be divisible by 2.".format(
-                        brain.brain_name, self.m_size
-                    )
-                )
         self._initialize_tensorflow_references()
         self.load = load
 
@@ -160,11 +148,7 @@ def _initialize_graph(self):
     def _load_graph(self, model_path: str, reset_global_steps: bool = False) -> None:
         with self.graph.as_default():
             self.saver = tf.train.Saver(max_to_keep=self.keep_checkpoints)
-            logger.info(
-                "Loading model for brain {} from {}.".format(
-                    self.brain.brain_name, model_path
-                )
-            )
+            logger.info(f"Loading model from {model_path}.")
             ckpt = tf.train.get_checkpoint_state(model_path)
             if ckpt is None:
                 raise UnityPolicyException(

ml-agents/mlagents/trainers/settings.py

@@ -51,10 +51,21 @@ def as_dict(self):
 
 @attr.s(auto_attribs=True)
 class NetworkSettings:
-    @attr.s(auto_attribs=True)
+    @attr.s
     class MemorySettings:
-        sequence_length: int = 64
-        memory_size: int = 128
+        sequence_length: int = attr.ib(default=64)
+        memory_size: int = attr.ib(default=128)
+
+        @memory_size.validator
+        def _check_valid_memory_size(self, attribute, value):
+            if value <= 0:
+                raise TrainerConfigError(
+                    "When using a recurrent network, memory size must be greater than 0."
+                )
+            elif value % 2 != 0:
+                raise TrainerConfigError(
+                    "When using a recurrent network, memory size must be divisible by 2."
+                )
 
     normalize: bool = False
     hidden_units: int = 128

ml-agents/mlagents/trainers/tests/test_settings.py

@@ -6,6 +6,7 @@
 from mlagents.trainers.settings import (
     RunOptions,
     TrainerSettings,
+    NetworkSettings,
     PPOSettings,
     SACSettings,
     RewardSignalType,
@@ -155,6 +156,14 @@ def test_reward_signal_structure():
         )
 
 
+def test_memory_settings_validation():
+    with pytest.raises(TrainerConfigError):
+        NetworkSettings.MemorySettings(sequence_length=128, memory_size=63)
+
+    with pytest.raises(TrainerConfigError):
+        NetworkSettings.MemorySettings(sequence_length=128, memory_size=0)
+
+
 def test_parameter_randomization_structure():
     """
     Tests the ParameterRandomizationSettings structure method and all validators.

ml-agents/mlagents/trainers/trainer/trainer.py

@@ -121,7 +121,7 @@ def export_model(self, name_behavior_id: str) -> None:
         Exports the model
         """
         policy = self.get_policy(name_behavior_id)
-        settings = SerializationSettings(policy.model_path, policy.brain.brain_name)
+        settings = SerializationSettings(policy.model_path, self.brain_name)
         export_policy_model(settings, policy.graph, policy.sess)
 
     @abc.abstractmethod