def policy_improv(): policy_stable = True for i in xrange(max_car+1): for j in xrange(max_car+1): temp = pi[i, j] rewardlist = np.zeros(11) index = 0 for k in xrange(-max_move, max_move+1): rewardlist[index] = car_and_reward(i, j, k, V) index += 1 pi[i, j] = np.argmax(rewardlist)-5 if temp != pi[i, j]: policy_stable = False return policy_stable
def policy_improv():
policy_stable = True
for i in xrange(max_car + 1):
for j in xrange(max_car + 1):
temp = pi[i, j]
rewardlist = np.zeros(11)
index = 0
for k in xrange(-max_move, max_move + 1):
rewardlist[index] = car_and_reward(i, j, k, V)
index += 1
pi[i, j] = np.argmax(rewardlist) - 5
if temp != pi[i, j]:
policy_stable = False
return policy_stable
def policy_eval():
while True:
delta = 0
for i in xrange(max_car+1):
for j in xrange(max_car+1):
# print i,j
temp = V[i, j]
V[i, j] = car_and_reward(i, j, pi[i, j], V)
delta = max(delta, abs(temp-V[i, j]))
# print V
# raw_input('Enter')
# print V
# raw_input('enter')
if delta < 1e-9:
break
def policy_eval():
while True:
delta = 0
for i in xrange(max_car + 1):
for j in xrange(max_car + 1):
# print i,j
temp = V[i, j]
V[i, j] = car_and_reward(i, j, pi[i, j], V)
delta = max(delta, abs(temp - V[i, j]))
# print V
# raw_input('Enter')
# print V
# raw_input('enter')
if delta < 1e-9:
break
