new observation properties for testing of technical limitations

This commit is contained in:
Steffen Illium
2021-11-05 15:59:19 +01:00
parent b5c6105b7b
commit d69cf75c15
9 changed files with 424 additions and 263 deletions

@ -5,6 +5,7 @@ import numpy as np
import yaml
from environments import helpers as h
from environments.helpers import Constants as c
from environments.factory.factory_dirt import DirtFactory
from environments.factory.factory_dirt_item import DirtItemFactory
from environments.logging.recorder import RecorderCallback
@ -15,29 +16,30 @@ warnings.filterwarnings('ignore', category=UserWarning)
if __name__ == '__main__':
model_name = 'DQN_1631187073'
model_name = 'DQN_163519000'
run_id = 0
seed = 69
out_path = Path('debug_out/DQN_1635176929/0_DQN_1635176929')
n_agents = 2
out_path = Path('debug_out/DQN_163519000/1_DQN_163519000')
model_path = out_path
with (out_path / f'env_params.json').open('r') as f:
env_kwargs = yaml.load(f, Loader=yaml.FullLoader)
env_kwargs.update(additional_agent_placeholder=None, n_agents=4)
if gain_amount := env_kwargs.get('dirt_properties', {}).get('gain_amount', None):
env_kwargs['dirt_properties']['max_spawn_amount'] = gain_amount
del env_kwargs['dirt_properties']['gain_amount']
env_kwargs.update(additional_agent_placeholder=None, n_agents=n_agents)
if gain_amount := env_kwargs.get('dirt_prop', {}).get('gain_amount', None):
env_kwargs['dirt_prop']['max_spawn_amount'] = gain_amount
del env_kwargs['dirt_prop']['gain_amount']
env_kwargs.update(record_episodes=True)
env_kwargs.update(record_episodes=False)
this_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(4)]
models = [model_cls.load(this_model) for _ in range(n_agents)]
with RecorderCallback(filepath=Path() / 'recorder_out_DQN.json') as recorder:
# Init Env
with DirtItemFactory(**env_kwargs) as env:
with DirtFactory(**env_kwargs) as env:
obs_shape = env.observation_space.shape
# Evaluation Loop for i in range(n Episodes)
for episode in range(5):
@ -46,11 +48,11 @@ if __name__ == '__main__':
while not done_bool:
actions = [model.predict(
np.stack([env_state[i][j] for i in range(env_state.shape[0])]),
deterministic=True)[0] for j, model in enumerate(models)]
deterministic=False)[0] for j, model in enumerate(models)]
env_state, step_r, done_bool, info_obj = env.step(actions)
recorder.read_info(0, info_obj)
rew += step_r
# env.render()
env.render()
if done_bool:
recorder.read_done(0, done_bool)
break