def OldStuff():
tf.compat.v1.disable_eager_execution()
lr = 0.001
numGames = 10000
session = TriadGameSession()
observation = session.getState()
scores = []
agent = Agent(gamma=0.99,
lr=lr,
epsilon=1.0,
epsilonDec=0.0005,
inputSize=[len(observation)],
numActions=session.getMaxActions(),
memSize=1000000,
batchSize=1024)
for i in range(numGames):
done = False
score = 0
session = TriadGameSession()
observation = session.getState()
while not done:
action = agent.chooseAction(observation)
observationNext, reward, done = session.step(action)
score += reward
agent.store(observation, action, reward, observationNext, done)
observation = observationNext
agent.learn()
scores.append(score)
avgScore = np.mean(scores[-100:])
print('game:', i, 'score %.2f' % score, 'avgScore %.2f' % avgScore,
'epsilon %.2f' % agent.epsilon)
#agent.save()
print('Finished!')
score = 0
for i in range(num_games):
if i % 10 == 0 and i > 0:
avg_score = np.mean(scores[max(0, i - 10):(i + 1)])
print('episode: ', i, 'score: ', score,
' average score %.3f' % avg_score,
'epsilon %.3f' % brain.EPSILON)
else:
print('episode: ', i, 'score: ', score)
eps_history.append(brain.EPSILON)
done = False
observation = env.reset()
score = 0
while not done:
action = brain.chooseAction(observation)
observation_, reward, done, info = env.step(action)
score += reward
brain.storeTransition(observation, action, reward, observation_,
done)
observation = observation_
brain.learn()
scores.append(score)
for i in range(10):
done = False
observation = env.reset()
while not done:
action = brain.chooseAction(observation)
observation_, reward, done, info = env.step(action)
observation = session.getState()
scores = []
agent = Agent(gamma=0.99, lr=lr, epsilon=1.0, epsilonDec=0.0005,
inputSize=[len(observation)],
numActions=session.getMaxActions(),
memSize=1000000,
batchSize=64)
for i in range(numGames):
done = False
score = 0
session = TriadGameSession()
observation = session.getState()
while not done:
action = agent.chooseAction(observation)
observationNext, reward, done = session.step(action)
score += reward
agent.store(observation, action, reward, observationNext, done)
observation = observationNext
agent.learn()
scores.append(score)
avgScore = np.mean(scores[-100:])
print('game:', i,
'score %.2f' % score,
'avgScore %.2f' % avgScore,
'epsilon %.2f' % agent.epsilon)
agent.save()
print('Finished!')
