steffen/marl-factory-grid
1
0
Fork 0
mirror of https://github.com/illiumst/marl-factory-grid.git synced 2025-12-12 10:30:37 +01:00
Code Issues Packages Projects Releases Wiki Activity

Monitor and Recorder are Wrappers.

Browse Source
This commit is contained in:
Steffen Illium
2021-11-24 17:39:26 +01:00
parent 59484f49c9
commit b0d6c2e1ef
10 changed files with 241 additions and 350 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
reload_agent.py
View File

@@ -8,7 +8,7 @@ from environments import helpers as h
from environments.helpers import Constants as c
from environments.factory.factory_dirt import DirtFactory
from environments.factory.combined_factories import DirtItemFactory
from environments.logging.recorder import RecorderCallback
from environments.logging.recorder import EnvRecorder
warnings.filterwarnings('ignore', category=FutureWarning)
warnings.filterwarnings('ignore', category=UserWarning)
@@ -16,14 +16,13 @@ warnings.filterwarnings('ignore', category=UserWarning)
if __name__ == '__main__':
model_name = 'A2C_ItsDirt'
run_id = 0
determin = True
render=False
determin = False
render = True
record = True
seed = 67
n_agents = 1
out_path = Path('study_out/e_1_Now_with_doors/no_obs/dirt/A2C_Now_with_doors/0_A2C_Now_with_doors')
n_agents = 2
out_path = Path('study_out/e_1_obs_stack_3_gae_0.25_n_steps_16/seperate_N/dirt/A2C_obs_stack_3_gae_0.25_n_steps_16/0_A2C_obs_stack_3_gae_0.25_n_steps_16')
out_path_2 = Path('study_out/e_1_obs_stack_3_gae_0.25_n_steps_16/seperate_N/dirt/A2C_obs_stack_3_gae_0.25_n_steps_16/1_A2C_obs_stack_3_gae_0.25_n_steps_16')
model_path = out_path
with (out_path / f'env_params.json').open('r') as f:
@@ -33,42 +32,35 @@ if __name__ == '__main__':
env_kwargs['dirt_prop']['max_spawn_amount'] = gain_amount
del env_kwargs['dirt_prop']['gain_amount']
env_kwargs.update(record_episodes=record)
env_kwargs.update(record_episodes=record, done_at_collision=True)
this_model = out_path / 'model.zip'
other_model = out_path / 'model.zip'
model_cls = next(val for key, val in h.MODEL_MAP.items() if key in model_name)
models = [model_cls.load(this_model) for _ in range(n_agents)]
model_cls = next(val for key, val in h.MODEL_MAP.items() if key in out_path.parent.name)
models = [model_cls.load(this_model), model_cls.load(other_model)]
with RecorderCallback(filepath=Path() / 'recorder_out_DQN.json', occupation_map=True,
entities=['Agents']) as recorder:
# Init Env
with DirtFactory(**env_kwargs) as env:
obs_shape = env.observation_space.shape
# Evaluation Loop for i in range(n Episodes)
recorder.read_params(env.params)
for episode in range(200):
env_state = env.reset()
rew, done_bool = 0, False
while not done_bool:
if n_agents > 1:
actions = [model.predict(
np.stack([env_state[i][j] for i in range(env_state.shape[0])]),
deterministic=determin)[0] for j, model in enumerate(models)]
else:
actions = models[0].predict(env_state, deterministic=determin)[0]
if False:
if any([agent.pos in [door.pos for door in env.unwrapped[c.DOORS]]
for agent in env.unwrapped[c.AGENT]]):
print('On Door')
env_state, step_r, done_bool, info_obj = env.step(actions)
# Init Env
with DirtFactory(**env_kwargs) as env:
env = EnvRecorder(env)
obs_shape = env.observation_space.shape
# Evaluation Loop for i in range(n Episodes)
for episode in range(50):
env_state = env.reset()
rew, done_bool = 0, False
while not done_bool:
if n_agents > 1:
actions = [model.predict(
np.stack([env_state[i][j] for i in range(env_state.shape[0])]),
deterministic=determin)[0] for j, model in enumerate(models)]
else:
actions = models[0].predict(env_state, deterministic=determin)[0]
env_state, step_r, done_bool, info_obj = env.step(actions)
recorder.read_info(0, info_obj)
rew += step_r
if render:
env.render()
if done_bool:
recorder.read_done(0, done_bool)
break
rew += step_r
if render:
env.render()
if done_bool:
break
print(f'Factory run {episode} done, reward is:\n {rew}')
print('all done')