def test_soak(self):
state_shape = (50, 50, 6)
rm = ReplayMemory(self.sess,
buffer_size=10000,
state_shape=state_shape,
action_dim=2,
load_factor=1.5)
self.sess.run(tf.initialize_all_variables())
def s_for(i):
return np.random.random(state_shape)
import random
i = 0
for e in xrange(10000):
# add an episode to rm
episode_len = random.choice([5, 7, 9, 10, 15])
initial_state = s_for(i)
action_reward_state = []
for i in range(i + 1, i + episode_len + 1):
a, r, s2 = (i * 10) + 7, (i * 10) + 8, s_for(i)
action_reward_state.append((a, r, s2))
rm.add_episode(initial_state, action_reward_state)
i += episode_len + 1
# dump
print rm.current_stats()
# fetch a batch, of all items, but do nothing with it.
_ = rm.batch(idxs=range(10))
