def main():
sizeS = 9
sizeA = 4
tau = getTau()
rho = getRho()
gamma = 0.75
V, Q = valueIteration(sizeS, sizeA, tau, rho, gamma)
V1, Q1 = valueIteration(sizeS, sizeA, tau, rho, gamma, 1)
V2, Q2 = valueIteration(sizeS, sizeA, tau, rho, gamma, 2)
V3, Q3 = valueIteration(sizeS, sizeA, tau, rho, gamma, 3)
V4, Q4 = valueIteration(sizeS, sizeA, tau, rho, gamma, 4)
# print(V)
# print(V.reshape((3,3)))
plt.figure()
plt.subplot(3, 2, 6)
plt.imshow(V.reshape((3, 3)), cmap='hot')
plt.subplot(3, 2, 1)
plt.imshow(V1.reshape((3, 3)), cmap='hot')
plt.subplot(3, 2, 2)
plt.imshow(V2.reshape((3, 3)), cmap='hot')
plt.subplot(3, 2, 3)
plt.imshow(V3.reshape((3, 3)), cmap='hot')
plt.subplot(3, 2, 4)
plt.imshow(V4.reshape((3, 3)), cmap='hot')
plt.show()
def main():
print "~~~~~~~~~~ Value iteration rewards mean stop ~~~~~~~~~~\n\n"
temp = valueIteration.valueIteration()
valueIteration.STOP = 0
print "\n\n~~~~~~~~~~ Value iteration rewards mean nothing ~~~~~~~~~~\n\n"
valueIteration.valueIteration()
print "\n\n~~~~~~~~~~ Reinforcement Learning ~~~~~~~~~~\n\n"
reinforcementLearning(temp)
def main():
print "~~~~~~~~~~ Value iteration rewards mean stop ~~~~~~~~~~\n\n"
temp = valueIteration.valueIteration()
valueIteration.STOP = 0
print "\n\n~~~~~~~~~~ Value iteration rewards mean nothing ~~~~~~~~~~\n\n"
valueIteration.valueIteration()
print "\n\n~~~~~~~~~~ Reinforcement Learning ~~~~~~~~~~\n\n"
reinforcementLearning(temp)
def run_val_iter(tau, rho, gamma=GAMMA):
V = valueIteration(States.COUNT, Action.COUNT, tau, rho, gamma)
print('Converged V:')
print(V.reshape((3, 3)))
plt.subplots(2, 2)
plt.suptitle('Value Iteration -- Maze')
for iters in range(1, 5):
plt.subplot(2, 2, iters)
V = valueIteration(States.COUNT, Action.COUNT, tau, rho, gamma, iters)
plt.imshow(V.reshape((3, 3)), cmap='hot')
plt.title(str(iters) + ' iterations')
plt.axis('off')
plt.show()
def plot_graph(tau, rho):
gammas = np.arange(0.5, 0.99, 0.01)
V = np.zeros(gammas.shape)
for i, gamma in enumerate(gammas):
V[i] = valueIteration(States.COUNT, Action.COUNT, tau, rho, gamma)[-1]
plt.figure()
plt.plot(gammas, V)
plt.title(r'$s_0$ for different $\gamma$ values')
def q3():
gamma = 0.5
gamma_values = []
s_0 = []
for i in range(50):
gamma_values.append(gamma)
s_0.append(valueIteration.valueIteration(5, 2, tau, rho, gamma)[0])
gamma += 0.01
plt.plot(gamma_values, s_0, 'r-')
plt.ylabel('value of s_0')
plt.xlabel('gamma')
plt.show()
def main():
sizeS = 5
sizeA = 2
tau = getTau()
rho = getRho()
gamma = 0.75
V, Q = valueIteration(sizeS, sizeA, tau, rho, gamma)
V1, Q1 = valueIteration(sizeS, sizeA, tau, rho, 0.5)
V2, Q2 = valueIteration(sizeS, sizeA, tau, rho, 0.75)
V3, Q3 = valueIteration(sizeS, sizeA, tau, rho, 0.85)
all_gamma = [0.5 + (x / 100.0) for x in range(49)]
v_s0 = []
v_send = []
for gamma in all_gamma:
V, Q = valueIteration(sizeS, sizeA, tau, rho, gamma)
v_s0.append(V[4])
v_send.append(V[0])
# print(v_s0)
plt.figure()
plt.plot(all_gamma, v_s0, 'b', all_gamma, v_send, 'r')
plt.show()
def run_val_iter(tau, rho, gamma=GAMMA):
# V = valueIteration(States.COUNT, Action.COUNT,
# tau, rho, gamma)
# print('Converged V:')
# print(V)
plt.subplots(3, 1)
plt.suptitle('Value Iteration -- Patience, dear')
for i, gamma in enumerate([0.5, 0.75, 0.85]):
plt.subplot(3, 1, i + 1)
V = valueIteration(States.COUNT, Action.COUNT, tau, rho, gamma)
print('V for gamma = %f:' % gamma)
print(V)
plt.imshow(V.reshape((1, -1)), cmap='hot')
plt.title(r'$\gamma = %f$' % gamma)
plt.axis('off')
# print model reward function R(s,a)
print("\nREWARD FUNCTION :")
for s in game.states:
for a in game.actions:
print("start state = " + s + ", action = " + a + ", reward = " +
str(game.rewards(s, a)))
print("\n----------------------------")
print("END MDP model")
print("----------------------------\n")
# Run Value Iteration
print("\n----------------------------")
print("ITERATIONS OF MDP VALUE ITERATION")
print("----------------------------\n")
VI = valueIteration.valueIteration(game.states, game.actions, game.transitions,
game.rewards, epsilon, gamma)
# Run a complete episode from initial state to end state following the optimal policy
print("\n----------------------------")
print("OPTIMAL POLICY À PARTIR DE S0")
print("----------------------------\n")
valueIteration.playEpisode("s0", game.isEnd, VI, game.actions,
game.transitions, game.rewards, gamma)
print("\n----------------------------")
print("OPTIMAL POLICY À PARTIR DE S2")
print("----------------------------\n")
valueIteration.playEpisode("s2", game.isEnd, VI, game.actions,
game.transitions, game.rewards, gamma)
# Run Q Learning Iteration
print("\n----------------------------")
def plotHeapMap(iterations):
for i in range(1, iterations + 1):
V = valueIteration.valueIteration(9, 4, tau, rho, 0.75, i)
#print V
plt.imshow(V.reshape((3, 3)), cmap="hot")
plt.show()
def q1():
V = valueIteration.valueIteration(9, 4, tau, rho, 0.75)
print V
rho[i,j]=-6.0 #prob/reward of staying in same state if i==k: #if theres a wall/barrier if s.actions[a] == -1: tau[i,j,k]=1.0 else: tau[i,j,k]=0.2 else: if s.actions[a]==k: tau[i,j,k]=0.8 rho[i,j]=-2 V = valueIteration.valueIteration(9, 4, tau, rho, 0.75) plt.imshow(V.reshape((3, 3)), cmap="hot") plt.show() def plotHeapMap(iterations): for i in range(1,iterations+1): V = valueIteration.valueIteration(9, 4, tau, rho, 0.75, i) plt.imshow(V.reshape((3, 3)), cmap="hot") plt.show() #plotHeapMap(9)
action_tm = {
0: transM.tmJun2Jul,
1: transM.stmJun2Jul
} # transition matrices
else:
action_tm = {
0: transM.tmJul2Aug,
1: transM.stmJul2Aug
} # transition matrices
policy = []
if method_name == 'v':
print("\nValue Iteration")
value_iteration = valueIteration(states,
rewards,
actions,
action_tm,
DISCOUNT_FACTOR,
display_process=show_round)
policy = getActionNames(value_iteration.generate_policy(),
actions_names)
elif method_name == 'p':
print("\nPolicy Iteration")
policy_iteration = policyIteration(states,
rewards,
actions,
action_tm,
DISCOUNT_FACTOR,
display_process=show_round)
policy = getActionNames(policy_iteration.generate_policy(),
actions_names)
else:
def q2():
print(valueIteration.valueIteration(5, 2, tau, rho, 0.5))
print(valueIteration.valueIteration(5, 2, tau, rho, 0.75))
print(valueIteration.valueIteration(5, 2, tau, rho, 0.85))
def q1():
print(valueIteration.valueIteration(5, 2, tau, rho, 0.75))