Exemplos de MDPconverter.initIndex em Python

Exemplo n.º 1

Arquivo: model_pomdp.py Projeto: chongdashu/py-vgdl

def plotLSPIValues(gametype, layout, discountFactor=0.9, useTD=False, showValue=False):
    # build the game
    g = VGDLParser().parseGame(gametype)
    g.buildLevel(layout)
    
    # transform into an MDP and the mapping to observations
    C = MDPconverter(g)
    Ts, R, fMap = C.convert()    
    
    # find the the best least-squares approximation to the policy,
    # given only observations, not the state information
    if useTD:
        # state-based
        _, Tlspi = LSTD_PI_policy(fMap, Ts, R, discountFactor=discountFactor)
    else:
        # state-action-based
        _, Tlspi = LSPI_policy(fMap, Ts, R, discountFactor=discountFactor)
    
    # evaluate the policy
    Vlspi = trueValues(Tlspi, R, discountFactor=discountFactor)
        
    # plot those values    
    featurePlot((g.width, g.height), C.states, Vlspi)
    
    if showValue:
        # expected discounted reward at initial state
        Vinit = Vlspi[C.initIndex()]
        pylab.xlabel("V0=%.4f"%Vinit)

Exemplo n.º 2

def plotLSPIValues(gametype, layout, discountFactor=0.9, useTD=False, showValue=False):
    # build the game
    g = VGDLParser().parseGame(gametype)
    g.buildLevel(layout)
    
    # transform into an MDP and the mapping to observations
    C = MDPconverter(g)
    Ts, R, fMap = C.convert()    
    
    # find the the best least-squares approximation to the policy,
    # given only observations, not the state information
    if useTD:
        # state-based
        _, Tlspi = LSTD_PI_policy(fMap, Ts, R, discountFactor=discountFactor)
    else:
        # state-action-based
        _, Tlspi = LSPI_policy(fMap, Ts, R, discountFactor=discountFactor)
    
    # evaluate the policy
    Vlspi = trueValues(Tlspi, R, discountFactor=discountFactor)
        
    # plot those values    
    featurePlot((g.width, g.height), C.states, Vlspi)
    
    if showValue:
        # expected discounted reward at initial state
        Vinit = Vlspi[C.initIndex()]
        pylab.xlabel("V0=%.4f"%Vinit)

Exemplo n.º 3

Arquivo: model_mdp.py Projeto: lrhammond/sdsrl

def plotOptimalValues(gametype, layout, discountFactor=0.9, showValue=False):
    # build the game
    g = VGDLParser().parseGame(gametype)
    g.buildLevel(layout)

    # transform into an MDP
    C = MDPconverter(g)
    Ts, R, _ = C.convert()

    # find the optimal policy
    _, Topt = policyIteration(Ts, R, discountFactor=discountFactor)

    # evaluate the policy
    Vopt = trueValues(Topt, R, discountFactor=discountFactor)

    # plot those values
    featurePlot((g.width, g.height), C.states, Vopt, plotdirections=True)

    if showValue:
        # expected discounted reward at initial state
        Vinit = Vopt[C.initIndex()]
        pylab.xlabel("V0=%.4f" % Vinit)

Exemplo n.º 4

Arquivo: model_mdp.py Projeto: chongdashu/py-vgdl

def plotOptimalValues(gametype, layout, discountFactor=0.9, showValue=False):
    # build the game
    g = VGDLParser().parseGame(gametype)
    g.buildLevel(layout)
    
    # transform into an MDP
    C = MDPconverter(g)
    Ts, R, _ = C.convert()
    
    # find the optimal policy
    _, Topt = policyIteration(Ts, R, discountFactor=discountFactor)
    
    # evaluate the policy
    Vopt = trueValues(Topt, R, discountFactor=discountFactor)
        
    # plot those values    
    featurePlot((g.width, g.height), C.states, Vopt, plotdirections=True)
    
    if showValue:
        # expected discounted reward at initial state
        Vinit = Vopt[C.initIndex()]
        pylab.xlabel("V0=%.4f"%Vinit)