def test_move(self):
grid = ' #P\nG #'
gw = GridWorld(grid, move_value=-1, die_value=-20, win_value=10)
step_tests = [
# move into wall
((0,0), (1,0), (0,0), -1, False),
# move to free field
((0,0), (1,1), (1,1), -1, False),
# move to goal
((0,0), (0,1), (0,1), 10, True),
# die penalty
((0,0), (2,0), (2,0), -20, True),
# out of bounds #1
((0,0), (-1,0), (0,0), -1, False),
# out of bounds #1
((0,0), (10,0), (0,0), -1, False),
]
for start, to, end, reward, is_terminal in step_tests:
e, r, t = gw.move(start, to)
self.assertEqual(e, end)
self.assertEqual(r, reward)
self.assertEqual(t, is_terminal)
if __name__ == '__main__':
max_steps = 100
max_iters = 1000
seed = random.randint(0, 100)
agent = qAgent()
grid = GridWorld(size=8, force_fast=True, seed=seed)
grid.show()
print()
for iter in range(max_iters):
agent.set_grid(grid)
i, j = 0, 0 # initial state
cum_reward = 0
for step in range(max_steps):
action = agent.get_action(i, j)
new_i, new_j = grid.move(i, j, action)
reward, is_final = grid.get_reward(i, j)
cum_reward += reward
agent.update_q(i, j, new_i, new_j, action, reward)
if is_final:
break
i = new_i
j = new_j
if iter % 100 == 0:
print(
'Episode {} finished after {} steps with cumulative reward of {}'
.format(iter, step, cum_reward))
grid = GridWorld(size=8, force_fast=True, seed=seed)
print()
show_qtable(agent, grid.size)
print "currentstate:", currentstate_index, currentstate
# Posicion actual (considerando to do el vector de estado)
# Realmente la posicion son la columna 1 y 2
print "currentposition:", currentposition
# Se asegura de que el movimiento sea valido
validMove = False
while not validMove:
action_index = agent.getAction('qlearning')
action = agent.getActions()[action_index]
print "newaction:", action_index, action
# Recupera la nueva posicion a traves de ejecutar una accion
newposition = gridworld.move(
agent.getCurrentPosition()[-2:], action)
print "newposition:", newposition
if newposition:
validMove = True
print "+ valid move"
# Incluye el presupuesto en la posicion
newposition = np.append([agent.getBudgetState()], newposition)
print "newposition with budget:", newposition
# Calcula la recompensa que devuelve el ambiente
current_reward = reward.reward(currentposition, action,
newposition)
print "reward:", current_reward
