value_iter_pi = mdp.pi
plotter.plot_state_actions(value_iter_pi, rewards = grid.reward_states, sinks = grid.sink_states)
value_iter_data = np.zeros([TRIALS, ITER])
classic_q_data = np.zeros([TRIALS, ITER])
for t in range(TRIALS):
mdp.load_policy(filename='scen1.p')
q = QLearner(grid, mdp, moves=40)
r = 0.0
for i in range(ITER):
q.guide()
r = r + q.get_reward() / (ITER)
print "Value iter reward: " + str(r)
value_iter_data[t,:] = np.zeros(ITER) + r
r = 0.0
q.clear_states()
mdp.pi = QPolicy(q)
a = Analysis(W, H, ITER, rewards=rewards, sinks=sinks, desc='Q policy')
for i in range(ITER * SAMP):
q.rollout()
r = r + q.get_reward() / (ITER * SAMP)
print "Q learn reward: " + str(r)
if t == 0:
a.count_states(q.get_states())
a.show_states()
