def test_policy_iteration(self):
grid = self.get_sample_grid()
env = Environment(grid)
planner = PolicyIterationPlanner(env)
result = planner.plan()
print("Policy Iteration")
for r in result:
print(r)
def test_value_iteration(self):
grid = self.get_sample_grid()
env = Environment(grid)
planner = ValuteIterationPlanner(env)
result = planner.plan()
print("Value Iteration")
for r in result:
print(r)
def test_run_environment(self):
grid = self.get_sample_grid()
env = Environment(grid)
for i in range(100):
state = env.reset() # initialize agent position
self.assertEqual(state.row, len(env.grid) - 1)
self.assertEqual(state.column, 0)
goal = False
for t in range(10):
action = random.choice(env.action_space)
state, reward, done = env.step(action)
self.assertTrue(0 <= state.row < len(env.grid))
self.assertTrue(0 <= state.column < len(env.grid[0]))
if done:
print("Episode {}: get reward {}, {} timesteps".format(
i, reward, t + 1))
goal = True
break
if not goal:
print("Episode {}: no reward".format(i))
def post(self):
data = tornado.escape.json_decode(self.request.body)
grid = data["grid"]
plan_type = data["plan"]
move_prob = 0.8 # default value
try:
move_prob = float(data["prob"])
except ValueError:
pass
env = Environment(grid, move_prob=move_prob)
if plan_type == "value":
planner = ValuteIterationPlanner(env)
elif plan_type == "policy":
planner = PolicyIterationPlanner(env)
result = planner.plan()
planner.log.append(result)
self.write({"log": planner.log})