def main():
z_size=128
batch_size=32
learning_rate=0.0001
vae = CNN(z_size=z_size,
batch_size=batch_size,
learning_rate=learning_rate,
is_training=True,
reuse=False,
gpu_mode=False)
env = 'Breakout-v0'
done_reward = -1
gamma = 0.95
epsilon = 0.05
save_freq = 10000
batch_size = 32
memory_size = 2000
vae = vae
rnn = None
controller = None
memory = Memory(memory_size)
preprocessor = Preprocessor()
foresee_steps = 1
DQA = DQNAgent(env = env,
done_reward = done_reward,
gamma = gamma,
epsilon = epsilon,
save_freq = save_freq,
batch_size = batch_size,
memory = memory,
cnn = vae,
controller = controller,
preprocessor = preprocessor,
foresee_steps = foresee_steps,
#vae_path = 'tf_cnn/cnn.json',
#rnn_path = 'tf_rnn/rnn.json',
)
'''
tmp_reward = []
for _ in range(10):
_, reward = DQA.generate_path(is_random = False)
tmp_reward.append(reward)
print(np.mean(tmp_reward))
'''
'''
path, _ = DQA.generate_path(is_random = False)
DQA.test_vae(path)
#DQA.test_rnn(path)
#DQA.test_preprocess(path)
'''
o = open('results.txt','w')
o.write('Start!')
o.write('\n')
S = []
res = DQA.fill_memory(2000, is_random = True)
S.append(res)
DQA.train_vae()
DQA.train_rnn()
o.write(str(res))
o.write('\n')
o.close()
for i in range(100):
o = open('results.txt','a')
DQA.memory.clear()
res = DQA.fill_memory(300, is_random = False)
o.write(str(res))
o.write('\n')
o.close()
S.append(res)
DQA.train_vae()
DQA.train_rnn()
print(S)
'''
