Unity 机器学习代理工具包 (ML-Agents) 是一个开源项目,它使游戏和模拟能够作为训练智能代理的环境。
您最多选择25个主题 主题必须以中文或者字母或数字开头,可以包含连字符 (-),并且长度不得超过35个字符
 
 
 
 
 

39 行
1.1 KiB

from mlagents.trainers.tests.mock_brain import make_fake_trajectory
from mlagents.trainers.tests.dummy_config import create_sensor_specs_with_shapes
from mlagents_envs.base_env import ActionSpec
VEC_OBS_SIZE = 6
ACTION_SIZE = 4
def test_trajectory_to_agentbuffer():
length = 15
wanted_keys = [
"next_obs_0",
"next_obs_1",
"obs_0",
"obs_1",
"memory",
"masks",
"done",
"continuous_action",
"discrete_action",
"continuous_log_probs",
"discrete_log_probs",
"action_mask",
"prev_action",
"environment_rewards",
]
wanted_keys = set(wanted_keys)
trajectory = make_fake_trajectory(
length=length,
sensor_specs=create_sensor_specs_with_shapes([(VEC_OBS_SIZE,), (84, 84, 3)]),
action_spec=ActionSpec.create_continuous(ACTION_SIZE),
)
agentbuffer = trajectory.to_agentbuffer()
seen_keys = set()
for key, field in agentbuffer.items():
assert len(field) == length
seen_keys.add(key)
assert seen_keys == wanted_keys