A bazaar of sokokan datasets and solver
git clone https://github.com/koulanurag/sokoban-bazaar.git
cd sokoban-bazaar
pip install -e .- Train Datasets
from torch.utils.data import DataLoader
from sokoban_bazaar.dataset import get_dataset, pad_episodic_batch
episode_dataset, weighted_sampler = get_dataset(env_name="gym_sokoban:Sokoban-v2", dataset_name="expert")
episode_dataloader = DataLoader(
episode_dataset,
batch_size=256,
num_workers=0,
sampler=weighted_sampler,
collate_fn=pad_episodic_batch
)
for batch in episode_dataloader:
pass- Test Environments (in-development)
from sokoban_bazaar.dataset import get_test_envs
test_envs = get_test_envs(env_name="gym_sokoban:Sokoban-v2")
for env in test_envs:
done = False
episode_score = 0
while not done:
action = env.action_space.sample()
obs, reward, done, info = env.step(action)
episode_score += reward
print(f"Episode Score: {episode_score}| {'Success' if info['all_boxes_on_target'] else 'Failure'}")- Solver
import gym
import random
from sokoban_bazaar.solver import PDDL, symbolic_state
from sokoban_bazaar.utils import domain_pddl_path
# create env
env_name = "gym_sokoban:Sokoban-v2"
env = gym.make(env_name)
env.reset()
sym_state = symbolic_state(env.render(mode="tiny_rgb_array"))
# generate plan
pddl = PDDL(
sym_state,
domain_pddl_path=domain_pddl_path(env_name),
problem_name=f"task-{random.randint(0, int(1e+10))}",
domain_name="sokoban",
)
plan = pddl.search_plan()
# execute plan
for action in plan:
_, _, done, info = env.step(action)
if done:
break
print("Success" if info["all_boxes_on_target"] else "Failure")- Install:
pip install -e ".[test]" - Run:
pytest
If you have any questions or suggestions , you can contact me at koulanurag@gmail.com or open an issue on this GitHub repository.