Example #1
0
def testAugmented():
    from core import VGDLParser
    from pybrain.rl.experiments.episodic import EpisodicExperiment
    from mdpmap import MDPconverter
    from agents import PolicyDrivenAgent

    zelda_level2 = """
wwwwwwwwwwwww
wA wwk1ww   w
ww  ww    1 w
ww     wwww+w
wwwww1ww  www
wwwww  0  Gww
wwwwwwwwwwwww
"""

    from examples.gridphysics.mazes.rigidzelda import rigidzelda_game
    g = VGDLParser().parseGame(rigidzelda_game)
    g.buildLevel(zelda_level2)
    env = GameEnvironment(g,
                          visualize=False,
                          recordingEnabled=True,
                          actionDelay=150)
    C = MDPconverter(g, env=env, verbose=True)
    Ts, R, _ = C.convert()
    print C.states
    print Ts[0]
    print R
    env.reset()
    agent = PolicyDrivenAgent.buildOptimal(env)
    env.visualize = True
    env.reset()
    task = GameTask(env)
    exper = EpisodicExperiment(task, agent)
    exper.doEpisodes(1)
Example #2
0
def testAugmented():
    from core import VGDLParser
    from pybrain.rl.experiments.episodic import EpisodicExperiment
    from mdpmap import MDPconverter
    from agents import PolicyDrivenAgent    
    
    
    zelda_level2 = """
wwwwwwwwwwwww
wA wwk1ww   w
ww  ww    1 w
ww     wwww+w
wwwww1ww  www
wwwww  0  Gww
wwwwwwwwwwwww
"""

    
    from examples.gridphysics.mazes.rigidzelda import rigidzelda_game
    g = VGDLParser().parseGame(rigidzelda_game)
    g.buildLevel(zelda_level2)
    env = GameEnvironment(g, visualize=False,
                          recordingEnabled=True, actionDelay=150)
    C = MDPconverter(g, env=env, verbose=True)
    Ts, R, _ = C.convert()
    print C.states
    print Ts[0]
    print R
    env.reset()
    agent = PolicyDrivenAgent.buildOptimal(env)
    env.visualize = True
    env.reset()
    task = GameTask(env)    
    exper = EpisodicExperiment(task, agent)
    exper.doEpisodes(1)
Example #3
0
    def buildOptimal(game_env, discountFactor=0.99):
        """ Given a game, find the optimal (state-based) policy and 
        return an agent that is playing accordingly. """
        from mdpmap import MDPconverter
        C = MDPconverter(env=game_env)
        Ts, R, _ = C.convert()
        policy, _ = policyIteration(Ts, R, discountFactor=discountFactor)
        game_env.reset()

        def x(*_):
            s = game_env.getState()
            #print s
            i = C.states.index(s)
            return i
        #return PolicyDrivenAgent(policy, lambda *_: C.states.index(game_env.getState()))
        return PolicyDrivenAgent(policy, x)