浏览代码

add team reward field to agent and proto

/develop/teammanager
Ruo-Ping Dong 4 年前
当前提交
90c9280e
共有 5 个文件被更改,包括 71 次插入31 次删除
  1. 46
      com.unity.ml-agents/Runtime/Agent.cs
  2. 38
      com.unity.ml-agents/Runtime/Grpc/CommunicatorObjects/AgentInfo.cs
  3. 11
      ml-agents-envs/mlagents_envs/communicator_objects/agent_info_pb2.py
  4. 6
      ml-agents-envs/mlagents_envs/communicator_objects/agent_info_pb2.pyi
  5. 1
      protobuf-definitions/proto/mlagents_envs/communicator_objects/agent_info.proto

46
com.unity.ml-agents/Runtime/Agent.cs


public float reward;
/// <summary>
/// The current team reward received by the agent.
/// </summary>
public float teamReward;
/// <summary>
/// Whether the agent is done or not.
/// </summary>
public bool done;

/// Additionally, the magnitude of the reward should not exceed 1.0
float m_Reward;
/// Represents the team reward the agent accumulated during the current step.
float m_TeamReward;
/// Keeps track of the cumulative reward in this episode.
float m_CumulativeReward;

/// OnActionReceived method.
/// </summary>
float[] m_LegacyActionCache;
ITeamManager m_TeamManager;
ITeamManager m_TeamManager;

}
}
// Request the last decision with no callbacks
if (m_TeamManager != null)
{
// Send final observations to TeamManager if it exists.
// The TeamManager is responsible to keeping track of the Agent after it's
// done, including propagating any "posthumous" rewards.
m_TeamManager.OnAgentDone(this, doneReason, sensors);
}
else
{
SendDoneToTrainer();
}
// We request a decision so Python knows the Agent is done immediately
m_Brain?.RequestDecision(m_Info, sensors);
ResetSensors();
// We also have to write any to any DemonstationStores so that they get the "done" flag.
foreach (var demoWriter in DemonstrationWriters)

m_RequestAction = false;
m_RequestDecision = false;
m_Info.storedActions.Clear();
}
public void SendDoneToTrainer()
{
// We request a decision so Python knows the Agent is done immediately
m_Brain?.RequestDecision(m_Info, sensors);
ResetSensors();
}
/// <summary>

m_CumulativeReward += increment;
}
public void AddRewardAfterDeath(float increment)
public void SetTeamReward(float reward)
{
#if DEBUG
Utilities.DebugCheckNanAndInfinity(reward, nameof(reward), nameof(SetTeamReward));
#endif
m_TeamReward += reward;
}
public void AddTeamReward(float increment)
m_Info.reward += increment;
#if DEBUG
Utilities.DebugCheckNanAndInfinity(increment, nameof(increment), nameof(AddTeamReward));
#endif
m_TeamReward += increment;
}
/// <summary>

38
com.unity.ml-agents/Runtime/Grpc/CommunicatorObjects/AgentInfo.cs


string.Concat(
"CjNtbGFnZW50c19lbnZzL2NvbW11bmljYXRvcl9vYmplY3RzL2FnZW50X2lu",
"Zm8ucHJvdG8SFGNvbW11bmljYXRvcl9vYmplY3RzGjRtbGFnZW50c19lbnZz",
"L2NvbW11bmljYXRvcl9vYmplY3RzL29ic2VydmF0aW9uLnByb3RvIuoBCg5B",
"L2NvbW11bmljYXRvcl9vYmplY3RzL29ic2VydmF0aW9uLnByb3RvIv8BCg5B",
"X2lkGA4gASgFSgQIARACSgQIAhADSgQIAxAESgQIBBAFSgQIBRAGSgQIBhAH",
"SgQIDBANQiWqAiJVbml0eS5NTEFnZW50cy5Db21tdW5pY2F0b3JPYmplY3Rz",
"YgZwcm90bzM="));
"X2lkGA4gASgFEhMKC3RlYW1fcmV3YXJkGA8gASgCSgQIARACSgQIAhADSgQI",
"AxAESgQIBBAFSgQIBRAGSgQIBhAHSgQIDBANQiWqAiJVbml0eS5NTEFnZW50",
"cy5Db21tdW5pY2F0b3JPYmplY3RzYgZwcm90bzM="));
new pbr::GeneratedClrTypeInfo(typeof(global::Unity.MLAgents.CommunicatorObjects.AgentInfoProto), global::Unity.MLAgents.CommunicatorObjects.AgentInfoProto.Parser, new[]{ "Reward", "Done", "MaxStepReached", "Id", "ActionMask", "Observations", "TeamManagerId" }, null, null, null)
new pbr::GeneratedClrTypeInfo(typeof(global::Unity.MLAgents.CommunicatorObjects.AgentInfoProto), global::Unity.MLAgents.CommunicatorObjects.AgentInfoProto.Parser, new[]{ "Reward", "Done", "MaxStepReached", "Id", "ActionMask", "Observations", "TeamManagerId", "TeamReward" }, null, null, null)
}));
}
#endregion

actionMask_ = other.actionMask_.Clone();
observations_ = other.observations_.Clone();
teamManagerId_ = other.teamManagerId_;
teamReward_ = other.teamReward_;
_unknownFields = pb::UnknownFieldSet.Clone(other._unknownFields);
}

}
}
/// <summary>Field number for the "team_reward" field.</summary>
public const int TeamRewardFieldNumber = 15;
private float teamReward_;
[global::System.Diagnostics.DebuggerNonUserCodeAttribute]
public float TeamReward {
get { return teamReward_; }
set {
teamReward_ = value;
}
}
[global::System.Diagnostics.DebuggerNonUserCodeAttribute]
public override bool Equals(object other) {
return Equals(other as AgentInfoProto);

if(!actionMask_.Equals(other.actionMask_)) return false;
if(!observations_.Equals(other.observations_)) return false;
if (TeamManagerId != other.TeamManagerId) return false;
if (!pbc::ProtobufEqualityComparers.BitwiseSingleEqualityComparer.Equals(TeamReward, other.TeamReward)) return false;
return Equals(_unknownFields, other._unknownFields);
}

hash ^= actionMask_.GetHashCode();
hash ^= observations_.GetHashCode();
if (TeamManagerId != 0) hash ^= TeamManagerId.GetHashCode();
if (TeamReward != 0F) hash ^= pbc::ProtobufEqualityComparers.BitwiseSingleEqualityComparer.GetHashCode(TeamReward);
if (_unknownFields != null) {
hash ^= _unknownFields.GetHashCode();
}

output.WriteRawTag(112);
output.WriteInt32(TeamManagerId);
}
if (TeamReward != 0F) {
output.WriteRawTag(125);
output.WriteFloat(TeamReward);
}
if (_unknownFields != null) {
_unknownFields.WriteTo(output);
}

if (TeamManagerId != 0) {
size += 1 + pb::CodedOutputStream.ComputeInt32Size(TeamManagerId);
}
if (TeamReward != 0F) {
size += 1 + 4;
}
if (_unknownFields != null) {
size += _unknownFields.CalculateSize();
}

observations_.Add(other.observations_);
if (other.TeamManagerId != 0) {
TeamManagerId = other.TeamManagerId;
}
if (other.TeamReward != 0F) {
TeamReward = other.TeamReward;
}
_unknownFields = pb::UnknownFieldSet.MergeFrom(_unknownFields, other._unknownFields);
}

}
case 112: {
TeamManagerId = input.ReadInt32();
break;
}
case 125: {
TeamReward = input.ReadFloat();
break;
}
}

11
ml-agents-envs/mlagents_envs/communicator_objects/agent_info_pb2.py


name='mlagents_envs/communicator_objects/agent_info.proto',
package='communicator_objects',
syntax='proto3',
serialized_pb=_b('\n3mlagents_envs/communicator_objects/agent_info.proto\x12\x14\x63ommunicator_objects\x1a\x34mlagents_envs/communicator_objects/observation.proto\"\xea\x01\n\x0e\x41gentInfoProto\x12\x0e\n\x06reward\x18\x07 \x01(\x02\x12\x0c\n\x04\x64one\x18\x08 \x01(\x08\x12\x18\n\x10max_step_reached\x18\t \x01(\x08\x12\n\n\x02id\x18\n \x01(\x05\x12\x13\n\x0b\x61\x63tion_mask\x18\x0b \x03(\x08\x12<\n\x0cobservations\x18\r \x03(\x0b\x32&.communicator_objects.ObservationProto\x12\x17\n\x0fteam_manager_id\x18\x0e \x01(\x05J\x04\x08\x01\x10\x02J\x04\x08\x02\x10\x03J\x04\x08\x03\x10\x04J\x04\x08\x04\x10\x05J\x04\x08\x05\x10\x06J\x04\x08\x06\x10\x07J\x04\x08\x0c\x10\rB%\xaa\x02\"Unity.MLAgents.CommunicatorObjectsb\x06proto3')
serialized_pb=_b('\n3mlagents_envs/communicator_objects/agent_info.proto\x12\x14\x63ommunicator_objects\x1a\x34mlagents_envs/communicator_objects/observation.proto\"\xff\x01\n\x0e\x41gentInfoProto\x12\x0e\n\x06reward\x18\x07 \x01(\x02\x12\x0c\n\x04\x64one\x18\x08 \x01(\x08\x12\x18\n\x10max_step_reached\x18\t \x01(\x08\x12\n\n\x02id\x18\n \x01(\x05\x12\x13\n\x0b\x61\x63tion_mask\x18\x0b \x03(\x08\x12<\n\x0cobservations\x18\r \x03(\x0b\x32&.communicator_objects.ObservationProto\x12\x17\n\x0fteam_manager_id\x18\x0e \x01(\x05\x12\x13\n\x0bteam_reward\x18\x0f \x01(\x02J\x04\x08\x01\x10\x02J\x04\x08\x02\x10\x03J\x04\x08\x03\x10\x04J\x04\x08\x04\x10\x05J\x04\x08\x05\x10\x06J\x04\x08\x06\x10\x07J\x04\x08\x0c\x10\rB%\xaa\x02\"Unity.MLAgents.CommunicatorObjectsb\x06proto3')
,
dependencies=[mlagents__envs_dot_communicator__objects_dot_observation__pb2.DESCRIPTOR,])

message_type=None, enum_type=None, containing_type=None,
is_extension=False, extension_scope=None,
options=None, file=DESCRIPTOR),
_descriptor.FieldDescriptor(
name='team_reward', full_name='communicator_objects.AgentInfoProto.team_reward', index=7,
number=15, type=2, cpp_type=6, label=1,
has_default_value=False, default_value=float(0),
message_type=None, enum_type=None, containing_type=None,
is_extension=False, extension_scope=None,
options=None, file=DESCRIPTOR),
],
extensions=[
],

oneofs=[
],
serialized_start=132,
serialized_end=366,
serialized_end=387,
)
_AGENTINFOPROTO.fields_by_name['observations'].message_type = mlagents__envs_dot_communicator__objects_dot_observation__pb2._OBSERVATIONPROTO

6
ml-agents-envs/mlagents_envs/communicator_objects/agent_info_pb2.pyi


id = ... # type: builtin___int
action_mask = ... # type: google___protobuf___internal___containers___RepeatedScalarFieldContainer[builtin___bool]
team_manager_id = ... # type: builtin___int
team_reward = ... # type: builtin___float
@property
def observations(self) -> google___protobuf___internal___containers___RepeatedCompositeFieldContainer[mlagents_envs___communicator_objects___observation_pb2___ObservationProto]: ...

action_mask : typing___Optional[typing___Iterable[builtin___bool]] = None,
observations : typing___Optional[typing___Iterable[mlagents_envs___communicator_objects___observation_pb2___ObservationProto]] = None,
team_manager_id : typing___Optional[builtin___int] = None,
team_reward : typing___Optional[builtin___float] = None,
) -> None: ...
@classmethod
def FromString(cls, s: builtin___bytes) -> AgentInfoProto: ...

def ClearField(self, field_name: typing_extensions___Literal[u"action_mask",u"done",u"id",u"max_step_reached",u"observations",u"reward",u"team_manager_id"]) -> None: ...
def ClearField(self, field_name: typing_extensions___Literal[u"action_mask",u"done",u"id",u"max_step_reached",u"observations",u"reward",u"team_manager_id",u"team_reward"]) -> None: ...
def ClearField(self, field_name: typing_extensions___Literal[u"action_mask",b"action_mask",u"done",b"done",u"id",b"id",u"max_step_reached",b"max_step_reached",u"observations",b"observations",u"reward",b"reward",u"team_manager_id",b"team_manager_id"]) -> None: ...
def ClearField(self, field_name: typing_extensions___Literal[u"action_mask",b"action_mask",u"done",b"done",u"id",b"id",u"max_step_reached",b"max_step_reached",u"observations",b"observations",u"reward",b"reward",u"team_manager_id",b"team_manager_id",u"team_reward",b"team_reward"]) -> None: ...

1
protobuf-definitions/proto/mlagents_envs/communicator_objects/agent_info.proto


reserved 12; // deprecated CustomObservationProto custom_observation = 12;
repeated ObservationProto observations = 13;
int32 team_manager_id = 14;
float team_reward = 15;
}
正在加载...
取消
保存