def training():
print('___________TRAINING______________')
agents = []
for idx_done in range(no_iter):
agent = Agent(gamma=gamma,
epsilon=epsilon,
batch_size=batch_size,
n_actions=m,
eps_end=0.01,
input_dims=[m],
lr=lr)
agents.append(agent)
scores, eps_history = [], []
start_time = time.time()
results = []
trials = []
for trial in range(no_trials):
env = Environment(trial, status='train')
score = 0
done = False
observation = env.reset()
i = 0
for idx_done in range(no_iter):
i += 1
action = agents[idx_done].choose_action(observation)
observation_, reward, done, info = env.step(action)
score += reward
agents[idx_done].store_trainsition(observation, action, reward,
observation_, done)
agents[idx_done].learn()
observation = observation_
#print('epsilon %.2f' % agents[idx_done].epsilon)
result = env.CodeRunner(trial)
results.append(result)
scores.append(score)
#eps_history.append(agent.epsilon)
avg_score = np.mean(scores)
trials.append(trial)
print('episode', trial, 'score %.2f' % score,
'average score %.2f' % avg_score,
'epsilon %.4f' % agents[idx_done].epsilon)
mvn_score = moving_average(scores, 250)
plt.figure(figsize=(5, 5), dpi=100)
plt.plot(trials, scores, label='Episodic Average')
plt.plot(mvn_score, label='Moving Average')
plt.xlabel('Trials')
plt.ylabel('Reward')
plt.grid()
plt.title('Training trial vs reward')
trimester = time.strftime("_%Y_%m_%d-%H__%M_%S")
plt.savefig('./pic/TR_DQN_' + str(no_trials) + '_' + (trimester) + '.png')
print("--- %s TRAINING seconds ---" % (time.time() - start_time))
env.show_result(results)
handle_agents(opt='save', agents=agents)
def testing():
print('___________TESTING_____________')
agent = handle_agents(opt='load')
scores, eps_history = [], []
start_time = time.time()
results = []
trials = []
for trial in range(no_trials):
env = Environment(trial, status='test')
score = 0
done = False
observation = env.reset()
i = 0
while not done:
i += 1
action = agent.choose_action(observation)
observation_, reward, done, info = env.step(action)
score += reward
observation = observation_
result = env.CodeRunner(trial)
results.append(result)
scores.append(score)
avg_score = np.mean(scores[-100:])
trials.append(trial)
print('episode', trial, 'score %.2f' % score, 'average score %.2f' % avg_score,
'epsilon %.4f' % agent.epsilon)
mvn_score = moving_average(scores, 250)
plt.figure(figsize=(5, 5), dpi=100)
plt.plot(trials, scores, label='Episodic Average')
plt.plot(mvn_score, label='Moving Average')
plt.xlabel('Trials')
plt.ylabel('Reward')
plt.grid()
plt.title('Testing trial vs reward')
trimester = time.strftime("_%Y_%m_%d-%H__%M_%S")
plt.savefig('./pic/TstS_DQN_' + str(no_trials) + '_' + (trimester) + '.png')
print("--- %s seconds ---" % (time.time() - start_time))
env.show_result(results)
def training():
print('----------TRAIN---------')
print('Episode num: ', MAX_EPISODES)
start = time.time()
scores = []
episodes = []
results = []
scores_window = deque(maxlen=100)
eps = EPS_START
if env_name == 'BP':
actions = range(param.m)
state_size = param.m
action_size = param.m
dqn_agentS = []
if single_nn:
dqn_agent = DQNAgent(state_size, action_size, seed=0)
else:
for t in range(MAX_STEPS):
dqn_agent = DQNAgent(state_size, action_size, seed=0)
dqn_agentS.append(dqn_agent)
for episode in range(1, MAX_EPISODES + 1):
if env_name == 'BP':
env = Environment(episode, status='train', state_asnp=True)
state = env.reset()
score = 0
for t in range(MAX_STEPS):
if single_nn:
action = dqn_agent.act(state, eps)
else:
action = dqn_agentS[t].act(state, eps)
next_state, reward, done, info = env.step(action)
if single_nn:
dqn_agent.step(state, action, reward, next_state, done)
else:
dqn_agentS[t].step(state, action, reward, next_state, done)
state = next_state
score += reward
if done:
break
eps = max(eps * EPS_DECAY, EPS_MIN)
if episode % PRINT_EVERY == 0:
mean_score = np.mean(scores_window)
print('\r Progress {}/{}, average score:{:.2f}'.format(
episode, MAX_EPISODES, mean_score),
end="")
result = env.CodeRunner(episode)
results.append(result)
scores_window.append(score)
scores.append(score)
episodes.append(episode)
print('epi ', episode, ' score ', score, 'avg score ', np.mean(scores))
plt.figure(figsize=(10, 6))
plt.plot(episodes, scores)
plt.plot(pd.Series(scores).rolling(100).mean())
plt.title('DQN Training')
plt.xlabel('# of episodes')
plt.ylabel('score')
#plt.show()
plt.grid()
trimester = time.strftime("_%Y_%m_%d-%H__%M_%S")
plt.savefig('./pic/TR_DQN_' + str(MAX_EPISODES) + '_' + (trimester) +
'.png')
env.show_result(results, MAX_EPISODES)
print('TRAINING {} seconds'.format(time.time() - start))
print('----------TEST---------')
print('Episode num: ', MAX_EPISODES)
start = time.time()
scores = []
episodes = []
results = []
scores_window = deque(maxlen=100)
eps = 0
episode = 0
for episodeT in range(1, MAX_EPISODES + 1):
if env_name == 'BP':
env = Environment(episodeT, status='train', state_asnp=True)
actions = range(param.m)
state_size = param.m
action_size = param.m
state = env.reset()
score = 0
for tT in range(MAX_STEPS):
if single_nn:
action = dqn_agent.act(state, eps)
else:
action = dqn_agentS[tT].act(state, eps)
next_state, reward, done, info = env.step(action)
"""
if single_nn:
dqn_agent.step(state, action, reward, next_state, done)
else:
dqn_agentS[t].step(state, action, reward, next_state, done)
"""
state = next_state
score += reward
if done:
break
if episodeT % PRINT_EVERY == 0:
mean_score = np.mean(scores_window)
print('\r Progress {}/{}, average score:{:.2f}'.format(
episode, MAX_EPISODES, mean_score),
end="")
result = env.CodeRunner(episodeT)
results.append(result)
scores_window.append(score)
scores.append(score)
episodes.append(episodeT)
print('epi ', episodeT, ' score ', score, 'avg score ',
np.mean(scores))
plt.figure(figsize=(10, 6))
plt.plot(episodes, scores)
plt.plot(pd.Series(scores).rolling(100).mean())
plt.title('DQN Testing')
plt.xlabel('# of episodes')
plt.ylabel('score')
#plt.show()
plt.grid()
trimester = time.strftime("_%Y_%m_%d-%H__%M_%S")
plt.savefig('./pic/TS_DQN_' + str(MAX_EPISODES) + '_' + (trimester) +
'.png')
env.show_result(results, MAX_EPISODES)
print('TESTING {} seconds'.format(time.time() - start))
