Change BrainParametersProto to support ActionSpec

com.unity.ml-agents/Runtime/Communicator/GrpcExtensions.cs

@@ -96,14 +96,14 @@ public static BrainParametersProto ToProto(this BrainParameters bp, string name,
         {
             var brainParametersProto = new BrainParametersProto
             {
-                VectorActionSize = { bp.VectorActionSize },
-                VectorActionSpaceType = (SpaceTypeProto)bp.VectorActionSpaceType,
+                VectorActionSizeDeprecated = { bp.VectorActionSize },
+                VectorActionSpaceTypeDeprecated = (SpaceTypeProto)bp.VectorActionSpaceType,
                 BrainName = name,
                 IsTraining = isTraining
             };
             if (bp.VectorActionDescriptions != null)
             {
-                brainParametersProto.VectorActionDescriptions.AddRange(bp.VectorActionDescriptions);
+                brainParametersProto.VectorActionDescriptionsDeprecated.AddRange(bp.VectorActionDescriptions);
             }
             return brainParametersProto;
         }
@@ -126,13 +126,13 @@ public static BrainParametersProto ToBrainParametersProto(this ActionSpec action
             };
             if (actionSpec.NumContinuousActions > 0)
             {
-                brainParametersProto.VectorActionSize.Add(actionSpec.NumContinuousActions);
-                brainParametersProto.VectorActionSpaceType = SpaceTypeProto.Continuous;
+                brainParametersProto.VectorActionSizeDeprecated.Add(actionSpec.NumContinuousActions);
+                brainParametersProto.VectorActionSpaceTypeDeprecated = SpaceTypeProto.Continuous;
             }
             else if (actionSpec.NumDiscreteActions > 0)
             {
-                brainParametersProto.VectorActionSize.AddRange(actionSpec.BranchSizes);
-                brainParametersProto.VectorActionSpaceType = SpaceTypeProto.Discrete;
+                brainParametersProto.VectorActionSizeDeprecated.AddRange(actionSpec.BranchSizes);
+                brainParametersProto.VectorActionSpaceTypeDeprecated = SpaceTypeProto.Discrete;
             }
 
             // TODO handle ActionDescriptions?
@@ -148,9 +148,9 @@ public static BrainParameters ToBrainParameters(this BrainParametersProto bpp)
         {
             var bp = new BrainParameters
             {
-                VectorActionSize = bpp.VectorActionSize.ToArray(),
-                VectorActionDescriptions = bpp.VectorActionDescriptions.ToArray(),
-                VectorActionSpaceType = (SpaceType)bpp.VectorActionSpaceType
+                VectorActionSize = bpp.VectorActionSizeDeprecated.ToArray(),
+                VectorActionDescriptions = bpp.VectorActionDescriptionsDeprecated.ToArray(),
+                VectorActionSpaceType = (SpaceType)bpp.VectorActionSpaceTypeDeprecated
             };
             return bp;
         }

com.unity.ml-agents/Runtime/Grpc/CommunicatorObjects/Capabilities.cs

@@ -129,7 +129,7 @@ public bool CompressedChannelMapping {
     public const int HybridActionsFieldNumber = 4;
     private bool hybridActions_;
     /// <summary>
-    /// support for mixed (discrete + continuous) actions
+    /// support for hybrid action spaces (discrete + continuous)
     /// </summary>
     [global::System.Diagnostics.DebuggerNonUserCodeAttribute]
     public bool HybridActions {

ml-agents-envs/mlagents_envs/communicator_objects/brain_parameters_pb2.pyi

@@ -33,27 +33,56 @@ builtin___float = float
 builtin___int = int
 
 
+class ActionSpecProto(google___protobuf___message___Message):
+    DESCRIPTOR: google___protobuf___descriptor___Descriptor = ...
+    num_continuous_actions = ... # type: builtin___int
+    num_discrete_actions = ... # type: builtin___int
+    discrete_branch_sizes = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[builtin___int]
+    action_descriptions = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[typing___Text]
+
+    def __init__(self,
+        *,
+        num_continuous_actions : typing___Optional[builtin___int] = None,
+        num_discrete_actions : typing___Optional[builtin___int] = None,
+        discrete_branch_sizes : typing___Optional[typing___Iterable[builtin___int]] = None,
+        action_descriptions : typing___Optional[typing___Iterable[typing___Text]] = None,
+        ) -> None: ...
+    @classmethod
+    def FromString(cls, s: builtin___bytes) -> ActionSpecProto: ...
+    def MergeFrom(self, other_msg: google___protobuf___message___Message) -> None: ...
+    def CopyFrom(self, other_msg: google___protobuf___message___Message) -> None: ...
+    if sys.version_info >= (3,):
+        def ClearField(self, field_name: typing_extensions___Literal[u"action_descriptions",u"discrete_branch_sizes",u"num_continuous_actions",u"num_discrete_actions"]) -> None: ...
+    else:
+        def ClearField(self, field_name: typing_extensions___Literal[u"action_descriptions",b"action_descriptions",u"discrete_branch_sizes",b"discrete_branch_sizes",u"num_continuous_actions",b"num_continuous_actions",u"num_discrete_actions",b"num_discrete_actions"]) -> None: ...
+
 class BrainParametersProto(google___protobuf___message___Message):
     DESCRIPTOR: google___protobuf___descriptor___Descriptor = ...
-    vector_action_size = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[builtin___int]
-    vector_action_descriptions = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[typing___Text]
-    vector_action_space_type = ... # type: mlagents_envs___communicator_objects___space_type_pb2___SpaceTypeProto
+    vector_action_size_deprecated = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[builtin___int]
+    vector_action_descriptions_deprecated = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[typing___Text]
+    vector_action_space_type_deprecated = ... # type: mlagents_envs___communicator_objects___space_type_pb2___SpaceTypeProto
     brain_name = ... # type: typing___Text
     is_training = ... # type: builtin___bool
 
+    @property
+    def action_spec(self) -> ActionSpecProto: ...
+
     def __init__(self,
         *,
-        vector_action_size : typing___Optional[typing___Iterable[builtin___int]] = None,
-        vector_action_descriptions : typing___Optional[typing___Iterable[typing___Text]] = None,
-        vector_action_space_type : typing___Optional[mlagents_envs___communicator_objects___space_type_pb2___SpaceTypeProto] = None,
+        vector_action_size_deprecated : typing___Optional[typing___Iterable[builtin___int]] = None,
+        vector_action_descriptions_deprecated : typing___Optional[typing___Iterable[typing___Text]] = None,
+        vector_action_space_type_deprecated : typing___Optional[mlagents_envs___communicator_objects___space_type_pb2___SpaceTypeProto] = None,
         brain_name : typing___Optional[typing___Text] = None,
         is_training : typing___Optional[builtin___bool] = None,
+        action_spec : typing___Optional[ActionSpecProto] = None,
         ) -> None: ...
     @classmethod
     def FromString(cls, s: builtin___bytes) -> BrainParametersProto: ...
     def MergeFrom(self, other_msg: google___protobuf___message___Message) -> None: ...
     def CopyFrom(self, other_msg: google___protobuf___message___Message) -> None: ...
     if sys.version_info >= (3,):
-        def ClearField(self, field_name: typing_extensions___Literal[u"brain_name",u"is_training",u"vector_action_descriptions",u"vector_action_size",u"vector_action_space_type"]) -> None: ...
+        def HasField(self, field_name: typing_extensions___Literal[u"action_spec"]) -> builtin___bool: ...
+        def ClearField(self, field_name: typing_extensions___Literal[u"action_spec",u"brain_name",u"is_training",u"vector_action_descriptions_deprecated",u"vector_action_size_deprecated",u"vector_action_space_type_deprecated"]) -> None: ...
     else:
-        def ClearField(self, field_name: typing_extensions___Literal[u"brain_name",b"brain_name",u"is_training",b"is_training",u"vector_action_descriptions",b"vector_action_descriptions",u"vector_action_size",b"vector_action_size",u"vector_action_space_type",b"vector_action_space_type"]) -> None: ...
+        def HasField(self, field_name: typing_extensions___Literal[u"action_spec",b"action_spec"]) -> builtin___bool: ...
+        def ClearField(self, field_name: typing_extensions___Literal[u"action_spec",b"action_spec",u"brain_name",b"brain_name",u"is_training",b"is_training",u"vector_action_descriptions_deprecated",b"vector_action_descriptions_deprecated",u"vector_action_size_deprecated",b"vector_action_size_deprecated",u"vector_action_space_type_deprecated",b"vector_action_space_type_deprecated"]) -> None: ...

ml-agents-envs/mlagents_envs/mock_communicator.py

@@ -39,9 +39,11 @@ def __init__(
 
     def initialize(self, inputs: UnityInputProto) -> UnityOutputProto:
         bp = BrainParametersProto(
-            vector_action_size=[2],
-            vector_action_descriptions=["", ""],
-            vector_action_space_type=discrete if self.is_discrete else continuous,
+            vector_action_size_deprecated=[2],
+            vector_action_descriptions_deprecated=["", ""],
+            vector_action_space_type_deprecated=discrete
+            if self.is_discrete
+            else continuous,
             brain_name=self.brain_name,
             is_training=True,
         )

ml-agents-envs/mlagents_envs/rpc_utils.py

@@ -33,15 +33,15 @@ def behavior_spec_from_proto(
     observation_shape = [tuple(obs.shape) for obs in agent_info.observations]
     action_type = (
         ActionType.DISCRETE
-        if brain_param_proto.vector_action_space_type == 0
+        if brain_param_proto.vector_action_space_type_deprecated == 0
         else ActionType.CONTINUOUS
     )
     if action_type == ActionType.CONTINUOUS:
         action_shape: Union[
             int, Tuple[int, ...]
-        ] = brain_param_proto.vector_action_size[0]
+        ] = brain_param_proto.vector_action_size_deprecated[0]
     else:
-        action_shape = tuple(brain_param_proto.vector_action_size)
+        action_shape = tuple(brain_param_proto.vector_action_size_deprecated)
     return BehaviorSpec(observation_shape, action_type, action_shape)
 
 

ml-agents-envs/mlagents_envs/tests/test_rpc_utils.py

@@ -408,17 +408,17 @@ def test_action_masking_continuous():
 def test_agent_behavior_spec_from_proto():
     agent_proto = generate_list_agent_proto(1, [(3,), (4,)])[0]
     bp = BrainParametersProto()
-    bp.vector_action_size.extend([5, 4])
-    bp.vector_action_space_type = 0
+    bp.vector_action_size_deprecated.extend([5, 4])
+    bp.vector_action_space_type_deprecated = 0
     behavior_spec = behavior_spec_from_proto(bp, agent_proto)
     assert behavior_spec.is_action_discrete()
     assert not behavior_spec.is_action_continuous()
     assert behavior_spec.observation_shapes == [(3,), (4,)]
     assert behavior_spec.discrete_action_branches == (5, 4)
     assert behavior_spec.action_size == 2
     bp = BrainParametersProto()
-    bp.vector_action_size.extend([6])
-    bp.vector_action_space_type = 1
+    bp.vector_action_size_deprecated.extend([6])
+    bp.vector_action_space_type_deprecated = 1
     behavior_spec = behavior_spec_from_proto(bp, agent_proto)
     assert not behavior_spec.is_action_discrete()
     assert behavior_spec.is_action_continuous()

ml-agents/mlagents/trainers/tests/tensorflow/test_simple_rl.py

@@ -432,9 +432,11 @@ def record_demo(use_discrete, num_visual=0, num_vector=1):
         agent_info_protos = env.demonstration_protos[BRAIN_NAME]
         meta_data_proto = DemonstrationMetaProto()
         brain_param_proto = BrainParametersProto(
-            vector_action_size=[2] if use_discrete else [1],
-            vector_action_descriptions=[""],
-            vector_action_space_type=discrete if use_discrete else continuous,
+            vector_action_size_deprecated=[2] if use_discrete else [1],
+            vector_action_descriptions_deprecated=[""],
+            vector_action_space_type_deprecated=discrete
+            if use_discrete
+            else continuous,
             brain_name=BRAIN_NAME,
             is_training=True,
         )

ml-agents/mlagents/trainers/tests/torch/test_simple_rl.py

@@ -314,9 +314,11 @@ def record_demo(use_discrete, num_visual=0, num_vector=1):
         agent_info_protos = env.demonstration_protos[BRAIN_NAME]
         meta_data_proto = DemonstrationMetaProto()
         brain_param_proto = BrainParametersProto(
-            vector_action_size=[2] if use_discrete else [1],
-            vector_action_descriptions=[""],
-            vector_action_space_type=discrete if use_discrete else continuous,
+            vector_action_size_deprecated=[2] if use_discrete else [1],
+            vector_action_descriptions_deprecated=[""],
+            vector_action_space_type_deprecated=discrete
+            if use_discrete
+            else continuous,
             brain_name=BRAIN_NAME,
             is_training=True,
         )