def render(self):
"""Renders the Q-function and policy learned"""
if self.Q is None:
self.Q = self.V + np.swapaxes(self.C[0, :, :], 0, 1).reshape(
(self.actions.n_prim + self.actions.n_opt,
self.GridWorld.n_states))
gui.render_q(self.GridWorld, self.Q) # Need a way to include options
gui.render_policy(self.GridWorld,
self.policy) # Need a way to include options
################################################################################
print(env.state2coord)
print(env.coord2state)
print(env.state_actions)
for i, el in enumerate(env.state_actions):
print("s{}: {}".format(i, env.action_names[el]))
################################################################################
# Policy definition
# If you want to represent deterministic action you can just use the number of
# the action. Recall that in the terminal states only action 0 (right) is
# defined.
# In this case, you can use gui.renderpol to visualize the policy
################################################################################
pol = [1, 2, 0, 0, 1, 1, 0, 0, 0, 0, 3]
gui.render_policy(env, pol)
################################################################################
# Try to simulate a trajectory
# you can use env.step(s,a, render=True) to visualize the transition
################################################################################
env.render = True
state = 0
fps = 1
for i in range(5):
action = np.random.choice(env.state_actions[state])
nexts, reward, term = env.step(state,action)
state = nexts
time.sleep(1./fps)
################################################################################