def monte_carlo_demo():
np.random.seed(101)
env = SnakeEnv(10, [3, 6])
agent = ModelFreeAgent(env)
mc = MonteCarlo(0.5)
with timer('Timer Monte Carlo Iter'):
mc.monte_carlo_opt(agent, env)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
np.random.seed(101)
agent2 = TableAgent(env)
pi_algo = PolicyIteration()
with timer('Timer PolicyIter'):
pi_algo.policy_iteration(agent2)
print('return_pi={}'.format(eval_game(env, agent2)))
print(agent2.pi)
np.random.seed(101)
agent3 = ModelFreeAgent(env)
mc = SARSA(0.5)
with timer('Timer Monte Carlo Iter'):
mc.sarsa(agent3, env)
print('return_pi={}'.format(eval_game(env, agent3)))
print(agent3.pi)
def monte_carlo_demo():
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent2 = TableAgent(env)
pi_algo = PolicyIteration()
with timer('Timer PolicyIter'):
pi_algo.policy_iteration(agent2)
print('PolicyIteration:return_pi={}'.format(eval_game(env, agent2)))
print(agent2.pi)
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent3 = TableAgent(env)
vi_algo = ValueIteration()
vi_algo.value_iteration(agent3)
print('ValueIteration:return_pi={}'.format(eval_game(env, agent3)))
print(agent3.pi)
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent = ModelFreeAgent(env)
mc = MonteCarlo()
with timer('Timer Monte Carlo Iter'):
mc.monte_carlo_opt(agent, env)
print('MonteCarlo:return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
def test_easy():
np.random.seed(0)
sum_opt = 0
sum_0 = 0
sum_1 = 0
env = SnakeEnv(0, [3, 6])
for i in range(10000):
sum_opt += eval_game(env, policy_ref)
sum_0 += eval_game(env, policy_0)
sum_1 += eval_game(env, policy_1)
print('opt avg={}'.format(sum_opt / 10000.0))
print('0 avg={}'.format(sum_0 / 10000.0))
print('1 avg={}'.format(sum_1 / 10000.0))
def policy_iteration_demo2():
env = SnakeEnv(10, [3, 6])
agent = TableAgent(env)
agent.pi[:] = 0
print('return3={}'.format(eval_game(env, agent)))
agent.pi[:] = 1
print('return6={}'.format(eval_game(env, agent)))
agent.pi[97:100] = 0
print('return_ensemble={}'.format(eval_game(env, agent)))
pi_algo = PolicyIteration()
pi_algo.policy_iteration(agent)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
def first_easy():
sum_opt = 0
sum_0 = 0
sum_1 = 0
env = SnakeEnv(0, [3, 6])
countNum = 10000
for i in range(countNum):
sum_opt += eval_game(env, policy_ref)
sum_0 += eval_game(env, policy_0)
sum_1 += eval_game(env, policy_1)
print('policy_ref avg={}'.format(sum_opt / countNum))
print('policy_0 avg={}'.format(sum_0 / countNum))
print('policy_1 avg={}'.format(sum_1 / countNum))
def monte_carlo_demo():
env = SnakeEnv(10, [3, 6])
agent = ModelFreeAgent(env)
mc = MonteCarlo()
with timer('Timer Monte Carlo Iter'):
mc.monte_carlo_opt(agent, env)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
agent2 = TableAgent(env)
pi_algo = PolicyIteration()
with timer('Timer PolicyIter'):
pi_algo.policy_iteration(agent2)
print('return_pi={}'.format(eval_game(env, agent2)))
print(agent2.pi)
def policy_iteration_demo1():
env = SnakeEnv(0, [3, 6])
agent = TableAgent(env)
pi_algo = PolicyIteration()
pi_algo.policy_iteration(agent)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
def policy_iteration_demo1():
env = SnakeEnv(0, [3, 6]) #0代表不考虑梯子
agent = TableAgent(env) #表agent
pi_algo = PolicyIteration() #策略迭代模型
pi_algo.policy_iteration(agent) #获得新一时刻的状态值函数
print 'return_pi={}'.format(eval_game(env, agent))
print agent.pi
def value_iteration_demo():
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent = TableAgent(env)
vi_algo = ValueIteration()
vi_algo.value_iteration(agent)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
def policy_iteration_demo():
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent = TableAgent(env)
pi_algo = PolicyIterationWithTimer()
pi_algo.policy_iteration(agent)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
def generalized_iteration_demo():
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent = TableAgent(env)
pi_algo = GeneralizedPolicyIteration()
with timer('Timer GeneralizedIter'):
pi_algo.generalized_policy_iteration(agent)
print('return_pi={}'.format(eval_game(env, agent)))
def monte_carlo_demo2():
env = SnakeEnv(10, [3, 6])
agent = ModelFreeAgent(env)
mc = MonteCarlo(0.5)
with timer('Timer Monte Carlo Iter'):
mc.monte_carlo_opt(agent, env)
print('return_pi={}'.format(eval_game(env, agent)))
print(agent.pi)
def value_iteration_demo():
np.random.seed(0)
env = SnakeEnv(10, [3, 6])
agent = TableAgent(env)
pi_algo = ValueIteration()
with timer('Timer ValueIter'):
pi_algo.value_iteration(agent)
print 'return_pi={}'.format(eval_game(env, agent))
