def test():
try:
from gridworld.chainwalk import Chainwalk
except:
print("Unable to import Chainwalk for test!")
return
cw = Chainwalk()
trace = cw.trace()
zeros = np.zeros(cw.nfeatures())
w = LSPI(trace,0.0001,cw,zeros,show=True)
print(w)
def test():
try:
from gridworld.chainwalk import Chainwalk
except:
print "Unable to import Chainwalk for test!"
return
cw = Chainwalk()
trace = cw.trace()
zeros = np.zeros(cw.nfeatures())
w = LSPI(trace,0.0001,cw,zeros,show=True)
print w
gw.background()
gw.mainloop()
if test_sarsa:
gw = GridworldGui(nrows=9, ncols=9, endstates=[0], walls=[])
learner = Sarsa(8, 81, 0.5, 0.9, 0.9, 0.1)
learner.learn(10000, gw, verbose=True)
pi = [learner.best(s) for s in range(gw.nstates)]
gw.set_arrows(pi)
gw.background()
gw.mainloop()
if test_chainwalk:
cw = Chainwalk()
t = cw.trace(1000)
policy0 = np.zeros(cw.nfeatures())
print(LSTDQ(t, cw, policy0))
if test_scale:
gw = GridworldGui(nrows=64, ncols=64, size=8, endstates=[0], walls=[])
try:
t = pickle.load(open("scale_trace.pck"))
except:
t = gw.trace(100000) #, show = False)
pickle.dump(t, open("scale_trace.pck", "w"), pickle.HIGHEST_PROTOCOL)
policy0 = np.zeros(gw.nfeatures())
#w0, weights0 = LSPI(t, 0.005, gw, policy0, maxiter=1, method="alt", debug = False, timer = True, show=False, format="csr")
w0, weights0 = LSPI(t,
0.005,
gw,
gw.background()
gw.mainloop()
if test_sarsa:
gw = GridworldGui(nrows = 9, ncols = 9, endstates = [0], walls = [])
learner = Sarsa(8, 81, 0.5, 0.9, 0.9, 0.1)
learner.learn(10000, gw, verbose=True)
pi = [learner.best(s) for s in range(gw.nstates)]
gw.set_arrows(pi)
gw.background()
gw.mainloop()
if test_chainwalk:
cw = Chainwalk()
t = cw.trace(1000)
policy0 = np.zeros(cw.nfeatures())
print LSTDQ(t, cw, policy0)
if test_scale:
gw = GridworldGui(nrows=64,ncols=64, size=8, endstates = [0], walls=[])
try:
t = pickle.load(open("scale_trace.pck"))
except:
t = gw.trace(100000)#, show = False)
pickle.dump(t,open("scale_trace.pck","w"),pickle.HIGHEST_PROTOCOL)
policy0 = np.zeros(gw.nfeatures())
#w0, weights0 = LSPI(t, 0.005, gw, policy0, maxiter=1, method="alt", debug = False, timer = True, show=False, format="csr")
w0, weights0 = LSPI(t, 0.005, gw, policy0, maxiter=10, method="parallel", debug = False, timer = True, show=True,ncpus=6)
#w0, weights0 = LSPI(t, 0.005, gw, policy0, maxiter=10, method="sparse", debug = False, timer = True, show=True, format="csr")
pi = [gw.linear_policy(w0,s) for s in range(gw.nstates)]