def test_config_init(self, train_config):
c = train_config
config = RAINBOW.generate_config({})
config["frame_config"]["models"] = ["QNet", "QNet"]
config["frame_config"]["model_kwargs"] = [{
"state_dim": c.observe_dim,
"action_num": c.action_num
}] * 2
rainbow = RAINBOW.init_from_config(config)
old_state = state = t.zeros([1, c.observe_dim], dtype=t.float32)
action = t.zeros([1, 1], dtype=t.int)
rainbow.store_episode([{
"state": {
"state": old_state
},
"action": {
"action": action
},
"next_state": {
"state": state
},
"reward": 0,
"terminal": False,
} for _ in range(3)])
rainbow.update()
def rainbow_train(self, train_config):
c = train_config
# cpu is faster for testing full training.
q_net = smw(QNet(c.observe_dim, c.action_num), "cpu", "cpu")
q_net_t = smw(QNet(c.observe_dim, c.action_num), "cpu", "cpu")
rainbow = RAINBOW(q_net, q_net_t,
t.optim.Adam,
c.value_min,
c.value_max,
reward_future_steps=c.reward_future_steps,
replay_device="cpu",
replay_size=c.replay_size)
return rainbow
def rainbow(self, train_config, device, dtype):
c = train_config
q_net = smw(QNet(c.observe_dim, c.action_num)
.type(dtype).to(device), device, device)
q_net_t = smw(QNet(c.observe_dim, c.action_num)
.type(dtype).to(device), device, device)
rainbow = RAINBOW(q_net, q_net_t,
t.optim.Adam,
c.value_min,
c.value_max,
reward_future_steps=c.reward_future_steps,
replay_device="cpu",
replay_size=c.replay_size)
return rainbow
def rainbow_vis(self, train_config, device, dtype, tmpdir):
c = train_config
tmp_dir = tmpdir.make_numbered_dir()
q_net = smw(QNet(c.observe_dim, c.action_num)
.type(dtype).to(device), device, device)
q_net_t = smw(QNet(c.observe_dim, c.action_num)
.type(dtype).to(device), device, device)
rainbow = RAINBOW(q_net, q_net_t,
t.optim.Adam,
c.value_min,
c.value_max,
reward_future_steps=c.reward_future_steps,
replay_device="cpu",
replay_size=c.replay_size,
visualize=True,
visualize_dir=str(tmp_dir))
return rainbow