RL::Interface Class Reference

Implements an interface providing fundamental capabilities required for reinforcement learning (over HTTP). More...

#include <RLInterface.h>

Collaboration diagram for RL::Interface:

Public Member Functions
	Interface (const char *server_address)
void	TryApplyMessage ()
	Non-blocking call to process any pending messages from the RL client. More...

Private Member Functions
	NONCOPYABLE (Interface)
std::string	Step (std::vector< GameCommand > &&commands)
	Process commands, update the simulation by one turn. More...
std::string	Reset (ScenarioConfig &&scenario)
	Reset the game state according to scenario, cleaning up existing games if required. More...
std::string	Evaluate (std::string &&code)
	Evaluate JS code in the engine such as applying arbitrary modifiers. More...
std::vector< std::string >	GetTemplates (const std::vector< std::string > &names) const
bool	IsGameRunning () const
std::string	SendGameMessage (GameMessage &&msg)
	Internal helper. More...
bool	TryGetGameMessage (GameMessage &msg)
	Internal helper. More...
void	ApplyMessage (const GameMessage &msg)
	Process any pending messages from the RL client. More...
std::string	GetGameState () const

Static Private Member Functions
static void *	MgCallback (mg_event event, struct mg_connection *conn, const struct mg_request_info *request_info)
static std::string	GetRequestContent (struct mg_connection *conn)

Private Attributes
GameMessage	m_GameMessage
ScenarioConfig	m_ScenarioConfig
std::string	m_ReturnValue
bool	m_NeedsGameState = false
std::mutex	m_Lock
std::mutex	m_MsgLock
std::condition_variable	m_MsgApplied
std::string	m_Code

Detailed Description

Implements an interface providing fundamental capabilities required for reinforcement learning (over HTTP).

This consists of enabling an external script to configure the scenario (via Reset) and then step the game engine manually and apply player actions (via Step). The interface also supports querying unit templates to provide information about max health and other potentially relevant game state information.

See source/tools/rlclient/ for the external client code.

The HTTP server is threaded. Flow of data (with the interface active): 0. The game/main thread calls TryApplyMessage()

• If no messages are pending, GOTO 0 (the simulation is not advanced).

TryApplyMessage locks m_MsgLock, pulls the message, processes it, advances the simulation, and sets m_ReturnValue.

1. TryApplyMessage notifies the RL thread that it can carry on and unlocks m_MsgLock. The main thread carries on frame rendering and goes back to 0.
2. The RL thread locks m_MsgLock, reads m_ReturnValue, unlocks m_MsgLock, and sends the gamestate as HTTP Response to the RL client.
3. The client processes the response and ultimately sends a new HTTP message to the RL Interface.
4. The RL thread locks m_MsgLock, pushes the message, and starts waiting on the game/main thread to notify it (step 2).
   • GOTO 0.

Constructor & Destructor Documentation

Interface()

RL::Interface::Interface

(

const char *

server_address

)

Member Function Documentation

ApplyMessage()

void RL::Interface::ApplyMessage ( const GameMessage &  msg )

private

Process any pending messages from the RL client.

Updates m_ReturnValue to the gamestate after messages have been processed.

Evaluate()

std::string RL::Interface::Evaluate ( std::string &&  code )

private

Evaluate JS code in the engine such as applying arbitrary modifiers.

Returns

the gamestate after script evaluation.

GetGameState()

std::string RL::Interface::GetGameState ( ) const

private

Returns

the full gamestate as a JSON strong. This uses the AI representation since it is readily available in the JS Engine.

GetRequestContent()

std::string RL::Interface::GetRequestContent ( struct mg_connection *  conn )

staticprivate

GetTemplates()

std::vector< std::string > RL::Interface::GetTemplates ( const std::vector< std::string > &  names ) const

private

Returns

template data for all templates of

Parameters

names.

IsGameRunning()

bool RL::Interface::IsGameRunning ( ) const

private

Returns

true if a game is currently running.

MgCallback()

void * RL::Interface::MgCallback ( mg_event  event, struct mg_connection *  conn, const struct mg_request_info *  request_info  )

staticprivate

NONCOPYABLE()

RL::Interface::NONCOPYABLE ( Interface  )

private

Reset()

std::string RL::Interface::Reset ( ScenarioConfig &&  scenario )

private

Reset the game state according to scenario, cleaning up existing games if required.

Returns

the gamestate after resetting.

SendGameMessage()

std::string RL::Interface::SendGameMessage ( GameMessage &&  msg )

private

Internal helper.

Move

Parameters

msg	into m_GameMessage, wait until it has been processed by the main thread, and

Returns

the gamestate after that message is processed. It is invalid to call this if m_GameMessage is not currently empty.

Step()

std::string RL::Interface::Step ( std::vector< GameCommand > &&  commands )

private

Process commands, update the simulation by one turn.

Returns

the gamestate after processing commands.

TryApplyMessage()

void RL::Interface::TryApplyMessage

(

)

Non-blocking call to process any pending messages from the RL client.

Updates m_ReturnValue to the gamestate after messages have been processed.

TryGetGameMessage()

bool RL::Interface::TryGetGameMessage ( GameMessage &  msg )

private

Internal helper.

Returns

true if m_GameMessage is not empty, and updates

Parameters

msg,false

otherwise (msg is then unchanged).

Member Data Documentation

m_Code

std::string RL::Interface::m_Code

private

m_GameMessage

GameMessage RL::Interface::m_GameMessage

private

m_Lock

std::mutex RL::Interface::m_Lock

mutableprivate

m_MsgApplied

std::condition_variable RL::Interface::m_MsgApplied

private

m_MsgLock

std::mutex RL::Interface::m_MsgLock

private

m_NeedsGameState

bool RL::Interface::m_NeedsGameState = false

private

m_ReturnValue

std::string RL::Interface::m_ReturnValue

private

m_ScenarioConfig

ScenarioConfig RL::Interface::m_ScenarioConfig

private

---

The documentation for this class was generated from the following files:

• /zpool0/docker-engine-docs/source/rlinterface/RLInterface.h
• /zpool0/docker-engine-docs/source/rlinterface/RLInterface.cpp