Individual Rewards

This commit is contained in:
Steffen Illium
2021-11-16 12:14:11 +01:00
parent b6bda84033
commit 0fe90f3ac0
11 changed files with 130 additions and 108 deletions

View File

@@ -18,14 +18,15 @@ if __name__ == '__main__':
model_name = 'A2C_ItsDirt'
run_id = 0
determin = True
seed = 67
n_agents = 1
out_path = Path('study_out/e_1_ItsDirt/no_obs/dirt/A2C_ItsDirt/0_A2C_ItsDirt')
out_path = Path('study_out/e_1_Now_with_doors/no_obs/dirt/A2C_Now_with_doors/0_A2C_Now_with_doors')
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=n_agents)
env_kwargs.update(additional_agent_placeholder=None, n_agents=n_agents, max_steps=150)
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']
@@ -49,9 +50,9 @@ if __name__ == '__main__':
if n_agents > 1:
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=determin)[0] for j, model in enumerate(models)]
else:
actions = models[0].predict(env_state, deterministic=True)[0]
actions = models[0].predict(env_state, deterministic=determin)[0]
if any([agent.pos in [door.pos for door in env.unwrapped[c.DOORS]]
for agent in env.unwrapped[c.AGENT]]):
print('On Door')