RL: q_learning_mountaincar.py

Keywords:

on_policy(SARSA)、off_policy(Q-Learning)、iter_max、t_max、gamma、eps、epsilon-greedy、

q_learning_mountaincar.py

"""
Model-free Prediction and Control
Example of SARSA Learning (on-policy) and Q-Learning (off-policy) using OpenAI gym MountainCar enviornment (https://gym.openai.com/envs/MountainCar-v0/)
Bolei Zhou for IERG6130, with parts of code adapted from Moustafa Alzantot (malzantot@ucla.edu)
"""
import numpy as np

import gym
from gym import wrappers

off_policy = True # if True use off-policy q-learning update, if False, use on-policy SARSA update

n_states = 40
iter_max = 5000

initial_lr = 1.0 # Learning rate
min_lr = 0.003
gamma = 1.0
t_max = 10000
eps = 0.1

def run_episode(env, policy=None, render=False):
    obs = env.reset()
    total_reward = 0
    step_idx = 0
    for _ in range(t_max):
        if render:
            env.render()
        if policy is None:
            action = env.action_space.sample()
        else:
            a,b = obs_to_state(env, obs)
            action = policy[a][b]
        obs, reward, done, _ = env.step(action)
        total_reward += gamma ** step_idx * reward
        step_idx += 1
        if done:
            break
    return total_reward

def obs_to_state(env, obs):
    """ Maps an observation to state """
    # we quantify the continous state space into discrete space
    env_low = env.observation_space.low
    env_high = env.observation_space.high
    env_dx = (env_high - env_low) / n_states
    a = int((obs[0] - env_low[0])/env_dx[0])
    b = int((obs[1] - env_low[1])/env_dx[1])
    return a, b

if __name__ == '__main__':
    env_name = 'MountainCar-v0'
    env = gym.make(env_name)
    env.seed(0)
    np.random.seed(0)
    if off_policy == True:
        print ('----- using Q Learning -----')
    else:
        print('------ using SARSA Learning ---')

    q_table = np.zeros((n_states, n_states, 3))
    for i in range(iter_max):
        obs = env.reset()
        total_reward = 0
        ## eta: learning rate is decreased at each step
        eta = max(min_lr, initial_lr * (0.85 ** (i//100)))
        for j in range(t_max):
            a, b = obs_to_state(env, obs)
            if np.random.uniform(0, 1) < eps:
                action = np.random.choice(env.action_space.n)
            else:
                action = np.argmax(q_table[a][b])
            obs, reward, done, _ = env.step(action)
            total_reward += reward
            # update q table
            a_, b_ = obs_to_state(env, obs)
            if off_policy == True:
                # use q-learning update (off-policy learning)
                q_table[a][b][action] = q_table[a][b][action] + eta * (reward + gamma *  np.max(q_table[a_][b_]) - q_table[a][b][action])
            else:
                # use SARSA update (on-policy learning)
                # epsilon-greedy policy on Q again
                if np.random.uniform(0,1) < eps:
                    action_ = np.random.choice(env.action_space.n)
                else:
                    action_ = np.argmax(q_table[a_][b_])
                q_table[a][b][action] = q_table[a][b][action] + eta * (reward + gamma *  q_table[a_][b_][action_] - q_table[a][b][action])
            if done:
                break
        if i % 200 == 0:
            print('Iteration #%d -- Total reward = %d.' %(i+1, total_reward))
    solution_policy = np.argmax(q_table, axis=2)
    solution_policy_scores = [run_episode(env, solution_policy, False) for _ in range(100)]
    print("Average score of solution = ", np.mean(solution_policy_scores))
    # Animate it
    for _ in range(2):
        run_episode(env, solution_policy, True)
    env.close()


# Results:
python q_learning_mountaincar.py

----- using Q Learning -----
Iteration #1 -- Total reward = -200.
Iteration #201 -- Total reward = -200.
Iteration #401 -- Total reward = -200.
Iteration #601 -- Total reward = -200.
Iteration #801 -- Total reward = -200.
Iteration #1001 -- Total reward = -200.
Iteration #1201 -- Total reward = -200.
Iteration #1401 -- Total reward = -200.
Iteration #1601 -- Total reward = -200.
Iteration #1801 -- Total reward = -200.
Iteration #2001 -- Total reward = -161.
Iteration #2201 -- Total reward = -200.
Iteration #2401 -- Total reward = -200.
Iteration #2601 -- Total reward = -200.
Iteration #2801 -- Total reward = -200.
Iteration #3001 -- Total reward = -153.
Iteration #3201 -- Total reward = -159.
Iteration #3401 -- Total reward = -200.
Iteration #3601 -- Total reward = -200.
Iteration #3801 -- Total reward = -200.
Iteration #4001 -- Total reward = -200.
Iteration #4201 -- Total reward = -200.
Iteration #4401 -- Total reward = -200.
Iteration #4601 -- Total reward = -200.
Iteration #4801 -- Total reward = -200.
Average score of solution =  -149.42


------ using SARSA Learning ---
Iteration #1 -- Total reward = -200.
Iteration #201 -- Total reward = -200.
Iteration #401 -- Total reward = -200.
Iteration #601 -- Total reward = -200.
Iteration #801 -- Total reward = -200.
Iteration #1001 -- Total reward = -200.
Iteration #1201 -- Total reward = -158.
Iteration #1401 -- Total reward = -160.
Iteration #1601 -- Total reward = -166.
Iteration #1801 -- Total reward = -200.
Iteration #2001 -- Total reward = -200.
Iteration #2201 -- Total reward = -191.
Iteration #2401 -- Total reward = -200.
Iteration #2601 -- Total reward = -154.
Iteration #2801 -- Total reward = -157.
Iteration #3001 -- Total reward = -200.
Iteration #3201 -- Total reward = -155.
Iteration #3401 -- Total reward = -159.
Iteration #3601 -- Total reward = -158.
Iteration #3801 -- Total reward = -194.
Iteration #4001 -- Total reward = -200.
Iteration #4201 -- Total reward = -200.
Iteration #4401 -- Total reward = -200.
Iteration #4601 -- Total reward = -158.
Iteration #4801 -- Total reward = -200.
Average score of solution =  -181.78

At a glance:

Q Learning SARSA