def test_per_nstep(self):
"""
PrioritizedReplayBuffer.on_episode_end() ignores Exception
Ref: https://gitlab.com/ymd_h/cpprb/-/issues/111
"""
rb = PrioritizedReplayBuffer(32, {
"rew": {},
"done": {}
},
Nstep={
"size": 4,
"rew": "rew",
"gamma": 0.5
})
for _ in range(10):
rb.add(rew=0.5, done=0.0)
rb.add(rew=0.5, done=1.0)
rb.on_episode_end()
s = rb.sample(16)
self.assertIn("discounts", s)
def test_save_cache_with_stack_compress(self):
rb = PrioritizedReplayBuffer(32,
env_dict={
'done': {
'dtype': 'bool'
},
'a': {
'shape': (3)
}
},
stack_compress='a')
a = np.array([0, 1, 2])
for i in range(3):
done = i == 2
rb.add(a=a, done=done)
if done:
rb.on_episode_end()
a += 1
rb.add(a=np.ones(3), done=False)
a_ = rb.get_all_transitions()["a"]
np.testing.assert_allclose(
a_,
np.asarray([[0., 1., 2.], [1., 2., 3.], [2., 3., 4.], [1., 1.,
1.]]))
observation = env.reset()
# Warming up
for n_step in range(100):
action = env.action_space.sample() # Random Action
next_observation, reward, done, info = env.step(action)
rb.add(obs=observation,
act=action,
rew=reward,
next_obs=next_observation,
done=done)
observation = next_observation
if done:
observation = env.reset()
rb.on_episode_end()
n_episode = 0
observation = env.reset()
for n_step in range(N_iteration):
if np.random.rand() < egreedy:
action = env.action_space.sample()
else:
Q = tf.squeeze(model(observation.reshape(1, -1)))
action = np.argmax(Q)
next_observation, reward, done, info = env.step(action)
rb.add(obs=observation,
act=action,
rew=reward,