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
110 行
3.4 KiB
110 行
3.4 KiB
import unittest.mock as mock
|
|
import pytest
|
|
import yaml
|
|
import mlagents.trainers.tests.mock_brain as mb
|
|
import numpy as np
|
|
from mlagents.trainers.rl_trainer import RLTrainer
|
|
from mlagents.trainers.tests.test_buffer import construct_fake_processing_buffer
|
|
from mlagents.trainers.buffer import AgentBuffer
|
|
|
|
|
|
@pytest.fixture
|
|
def dummy_config():
|
|
return yaml.safe_load(
|
|
"""
|
|
summary_path: "test/"
|
|
reward_signals:
|
|
extrinsic:
|
|
strength: 1.0
|
|
gamma: 0.99
|
|
"""
|
|
)
|
|
|
|
|
|
def create_mock_brain():
|
|
mock_brain = mb.create_mock_brainparams(
|
|
vector_action_space_type="continuous",
|
|
vector_action_space_size=[2],
|
|
vector_observation_space_size=8,
|
|
number_visual_observations=1,
|
|
)
|
|
return mock_brain
|
|
|
|
|
|
def create_rl_trainer():
|
|
mock_brainparams = create_mock_brain()
|
|
trainer = RLTrainer(mock_brainparams.brain_name, dummy_config(), True, 0)
|
|
return trainer
|
|
|
|
|
|
def create_mock_all_brain_info(brain_info):
|
|
return {"MockBrain": brain_info}
|
|
|
|
|
|
def create_mock_policy():
|
|
mock_policy = mock.Mock()
|
|
mock_policy.reward_signals = {}
|
|
mock_policy.retrieve_memories.return_value = np.zeros((1, 1), dtype=np.float32)
|
|
mock_policy.retrieve_previous_action.return_value = np.zeros(
|
|
(1, 1), dtype=np.float32
|
|
)
|
|
return mock_policy
|
|
|
|
|
|
@mock.patch("mlagents.trainers.rl_trainer.RLTrainer.add_policy_outputs")
|
|
@mock.patch("mlagents.trainers.rl_trainer.RLTrainer.add_rewards_outputs")
|
|
@pytest.mark.parametrize("num_vis_obs", [0, 1, 2], ids=["vec", "1 viz", "2 viz"])
|
|
def test_rl_trainer(add_policy_outputs, add_rewards_outputs, num_vis_obs):
|
|
trainer = create_rl_trainer()
|
|
trainer.policy = create_mock_policy()
|
|
fake_id = "fake_behavior_id"
|
|
fake_action_outputs = {
|
|
"action": [0.1, 0.1],
|
|
"value_heads": {},
|
|
"entropy": np.array([1.0], dtype=np.float32),
|
|
"learning_rate": 1.0,
|
|
}
|
|
mock_braininfo = mb.create_mock_braininfo(
|
|
num_agents=2,
|
|
num_vector_observations=8,
|
|
num_vector_acts=2,
|
|
num_vis_observations=num_vis_obs,
|
|
)
|
|
trainer.add_experiences(
|
|
fake_id, mock_braininfo, mock_braininfo, fake_action_outputs
|
|
)
|
|
|
|
# Remove one of the agents
|
|
next_mock_braininfo = mb.create_mock_braininfo(
|
|
num_agents=1,
|
|
num_vector_observations=8,
|
|
num_vector_acts=2,
|
|
num_vis_observations=num_vis_obs,
|
|
)
|
|
brain_info = trainer.construct_curr_info(next_mock_braininfo)
|
|
|
|
# assert construct_curr_info worked properly
|
|
assert len(brain_info.agents) == 1
|
|
assert len(brain_info.visual_observations) == num_vis_obs
|
|
assert len(brain_info.vector_observations) == 1
|
|
|
|
# Test end episode
|
|
trainer.end_episode()
|
|
for agent_id in trainer.episode_steps:
|
|
assert trainer.episode_steps[agent_id] == 0
|
|
assert len(trainer.processing_buffer[agent_id]["action"]) == 0
|
|
for rewards in trainer.collected_rewards.values():
|
|
for agent_id in rewards:
|
|
assert rewards[agent_id] == 0
|
|
|
|
|
|
def test_clear_update_buffer():
|
|
trainer = create_rl_trainer()
|
|
trainer.processing_buffer = construct_fake_processing_buffer()
|
|
trainer.update_buffer = AgentBuffer()
|
|
trainer.processing_buffer.append_to_update_buffer(
|
|
trainer.update_buffer, 2, batch_size=None, training_length=2
|
|
)
|
|
trainer.clear_update_buffer()
|
|
for _, arr in trainer.update_buffer.items():
|
|
assert len(arr) == 0
|