phi,
theta0,
policy=policy,
normalize_phi=False,
mu_next=1000)
methods = []
alpha = 0.2
mu = 2
gtd = td.GTD(alpha=alpha, beta=mu * alpha, phi=phi)
gtd.name = r"GTD $\alpha$={} $\mu$={}".format(alpha, mu)
gtd.color = "r"
methods.append(gtd)
alpha, mu = 0.4, 0.5
gtd = td.GTD2(alpha=alpha, beta=mu * alpha, phi=phi)
gtd.name = r"GTD2 $\alpha$={} $\mu$={}".format(alpha, mu)
gtd.color = "orange"
methods.append(gtd)
alpha = td.RMalpha(0.03, .1)
lam = .0
td0 = td.LinearTDLambda(alpha=alpha, lam=lam, phi=phi, gamma=gamma)
td0.name = r"TD({}) $\alpha$={}".format(lam, alpha)
td0.color = "k"
methods.append(td0)
alpha = .004
lam = 1.
td0 = td.LinearTDLambda(alpha=alpha, lam=lam, phi=phi, gamma=gamma)
td0.name = r"TD({}) $\alpha$={}".format(lam, alpha)
import td
import examples
from task import LinearDiscreteValuePredictionTask
import numpy as np
import features
n = 14
n_feat = 4
mdp = examples.BoyanChain(n, n_feat)
phi = features.spikes(n_feat, n)
gamma = .95
p0 = np.zeros(n_feat)
task = LinearDiscreteValuePredictionTask(mdp, gamma, phi, p0)
# define the methods to examine
gtd2 = td.GTD2(alpha=0.5, beta=0.5, phi=phi)
gtd2.name = "GTD2"
gtd2.color = "#0F6E08"
gtd = td.GTD(alpha=0.5, beta=0.5, phi=phi)
gtd.name = "GTD"
gtd.color = "#6E086D"
methods = []
alpha = .5
mu = 2.
gtd = td.GTD(alpha=alpha, beta=mu * alpha, phi=phi)
gtd.name = r"GTD $\alpha$={} $\mu$={}".format(alpha, mu)
gtd.color = "#6E086D"
methods.append(gtd)
