def main():
print("Start Atari games")
environment_name = "PongNoFrameskip-v4"
env = make_env(environment_name)
best_score = -np.inf
load_checkpoint = False
n_games = 500
lr = 0.0001
epsilon = 1
gamma = 0.99
input_dims = env.observation_space.shape
n_actions = env.action_space.n
eps_min = 0.01
eps_dec = 5e-7
replace = 1000
algo = None
mem_size = 50000
batch_size = 32
chkpt_dir = "models/"
algo = "DeepQAgent"
agent = DeepQAgent(lr, n_actions, input_dims, chkpt_dir, epsilon, gamma,
mem_size, batch_size, eps_min, eps_dec, replace, algo,
environment_name)
if load_checkpoint:
agent.load_models()
fname = agent.algo + "_" + agent.env_name + '_lr' + str(
agent.lr) + "_" + str(n_games) + "_games"
figure_file = "plots/" + fname + ".png"
n_steps = 0
scores, eps_history, steps_array = [], [], []
for i in range(n_games):
done = False
score = 0
observation = env.reset()
while not done:
action = agent.get_action(observation)
new_observation, reward, done, info = env.step(action)
score += reward
if not load_checkpoint:
agent.store_transition(observation, action, reward,
new_observation, int(done))
agent.learn()
observation = new_observation
n_steps += 1
scores.append(score)
steps_array.append(n_steps)
avg_score = np.mean(scores[-100:])
print(
"episode ", i + 1, "score: ", score,
"average score %.1f best score %.1f epsilon %.2f" %
(avg_score, best_score, agent.epsilon), " steps ", n_steps)
if avg_score > best_score:
if not load_checkpoint:
agent.save_models()
best_score = avg_score
eps_history.append(agent.epsilon)
plot_learning_curve(steps_array, scores, eps_history, figure_file)
print("End Atari games")
n_episodes = 10000
win_pct_list = []
scores = []
eps_history = []
for i in range(n_episodes):
done = False
score = 0
s = env.reset()
done = False
while not done:
a = A.pick_action(s)
s_, r, done, info = env.step(a)
A.learn(s, a, r, s_)
score += r
s = s_
scores.append(score)
eps_history.append(A.eps)
if i % 100 == 0:
win_pct = np.mean(scores[-100:])
win_pct_list.append(win_pct)
if i % 1000 == 0:
print('episode', i, 'win pct %.2f' % win_pct, 'eps %2.f' % A.eps)
#plt.plot(win_pct_list)
#plt.show()
x = [i + 1 for i in range(n_episodes)]