def lightsout(type='digital', size=4):
import importlib
p = importlib.import_module('latplan.puzzles.lightsout_{}'.format(type))
p.setup()
ics = reservoir_sampling(
untuple(
dijkstra(tuple(np.full(size * size, -1)), steps,
lambda config: p.successors(config))), instances)
gcs = np.full((1, size * size), -1)
generate(p, ics, gcs)
def puzzle(type='mnist', width=3, height=3):
import importlib
p = importlib.import_module('latplan.puzzles.puzzle_{}'.format(type))
p.setup()
ics = reservoir_sampling(
untuple(
dijkstra(tuple(np.arange(width * height)), steps,
lambda config: p.successors(config, width, height))),
instances)
gcs = np.arange(width * height).reshape((1, width * height))
generate(p, ics, gcs, width, height)
def hanoi(disks=5, towers=3):
import latplan.puzzles.hanoi as p
p.setup()
ics = [
np.zeros(disks, dtype=int), *reservoir_sampling(
untuple(
dijkstra(tuple(np.full(disks, towers - 1, dtype=int)), steps,
lambda config: p.successors(config, disks, towers))),
instances - 1)
]
gcs = np.full((1, disks), towers - 1, dtype=int)
generate(p, ics, gcs, disks, towers)
def sokoban_layout(limit = 1000, egocentric = False, objects = True, stage=0, test=False):
assert objects
list = ["sokoban_layout",limit,
("egocentric" if egocentric else "global"),
("object" if objects else "global"),
stage,
("test" if test else "train"),]
path = os.path.join(latplan.__path__[0],"puzzles","-".join(map(str,list))+".npz")
import gym
import pddlgym
import imageio
pre_layouts = []
suc_layouts = []
if egocentric:
layout_mode = "egocentric_layout"
else:
layout_mode = "layout"
env = gym.make("PDDLEnvSokoban-v0" if not test else "PDDLEnvSokobanTest-v0")
env.fix_problem_index(stage)
init, _ = env.reset()
init_layout = env.render(mode=layout_mode)
# reachability analysis
player = (init_layout == pddlgym.rendering.sokoban.PLAYER)
wall = (init_layout == pddlgym.rendering.sokoban.WALL)
reachable = compute_reachability_sokoban(wall,player)
relevant = np.maximum(reachable, wall)
print(f"{wall.sum()} wall objects:")
print(wall)
print(f"{reachable.sum()} reachable objects:")
print(reachable)
print(f"{relevant.sum()} relevant objects:")
print(relevant)
relevant = relevant.reshape(-1)
def successor(obs):
env.set_state(obs)
for action in env.action_space.all_ground_literals(obs, valid_only=True):
env.set_state(obs)
obs2, _, _, _ = env.step(action)
yield obs2
max_g = 0
for obs, close_list in dijkstra(init, float("inf"), successor, include_nonleaf=True, limit=limit):
max_g = max(max_g,close_list[obs]["g"])
pobs = close_list[obs]["parent"]
if pobs is None:
continue
env.set_state(pobs)
pre_layouts.append(env.render(mode=layout_mode))
env.set_state(obs)
suc_layouts.append(env.render(mode=layout_mode))
pre_layouts = np.array(pre_layouts)
suc_layouts = np.array(suc_layouts)
print(pre_layouts.shape)
print("max",pre_layouts.max(),"min",pre_layouts.min())
B, H, W = pre_layouts.shape
pre_layouts = pre_layouts.reshape((B,H*W))
suc_layouts = suc_layouts.reshape((B,H*W))
# shuffling
random_indices = np.arange(len(pre_layouts))
nr.shuffle(random_indices)
pre_layouts = pre_layouts[random_indices]
suc_layouts = suc_layouts[random_indices]
tile = 16
bboxes = tiled_bboxes(B, H, W, tile)
if not egocentric:
pre_layouts = pre_layouts[:,relevant]
suc_layouts = suc_layouts[:,relevant]
bboxes = bboxes[:,relevant]
# make it into a one-hot repr
eye = np.eye(pddlgym.rendering.sokoban.NUM_OBJECTS)
# B, H, W, C
pre_classes = eye[pre_layouts]
suc_classes = eye[suc_layouts]
print(pre_classes.shape)
np.savez_compressed(path,pres=pre_classes,sucs=suc_classes,bbox=bboxes,picsize=[H*tile,W*tile,3],max_g=max_g)
def sokoban_image(limit = 1000, egocentric = False, objects = True, stage=0, test=False):
list = ["sokoban_image",limit,
("egocentric" if egocentric else "global"),
("object" if objects else "global"),
stage,
("test" if test else "train"),]
path = os.path.join(latplan.__path__[0],"puzzles","-".join(map(str,list))+".npz")
import gym
import pddlgym
import imageio
pre_images = []
suc_images = []
if egocentric:
image_mode = "egocentric_crisp"
else:
image_mode = "human_crisp"
env = gym.make("PDDLEnvSokoban-v0" if not test else "PDDLEnvSokobanTest-v0")
env.fix_problem_index(stage)
init, _ = env.reset()
init_layout = env.render(mode="layout")
# reachability analysis
player = (init_layout == pddlgym.rendering.sokoban.PLAYER)
wall = (init_layout == pddlgym.rendering.sokoban.WALL)
reachable = compute_reachability_sokoban(wall,player)
relevant = np.maximum(reachable, wall)
print(f"{wall.sum()} wall objects:")
print(wall)
print(f"{reachable.sum()} reachable objects:")
print(reachable)
print(f"{relevant.sum()} relevant objects:")
print(relevant)
relevant = relevant.reshape(-1)
def successor(obs):
env.set_state(obs)
for action in env.action_space.all_ground_literals(obs, valid_only=True):
env.set_state(obs)
obs2, _, _, _ = env.step(action)
yield obs2
pairs = []
max_g = 0
for obs, close_list in dijkstra(init, float("inf"), successor, include_nonleaf=True, limit=limit):
max_g = max(max_g,close_list[obs]["g"])
pobs = close_list[obs]["parent"]
if pobs is None:
continue
pairs.append((pobs,obs))
threads = 16
pairss = []
len_per_thread = 1+(len(pairs) // threads)
for i in range(threads):
pairss.append(pairs[i*len_per_thread:(i+1)*len_per_thread])
from multiprocessing import Pool
with Pool(threads) as p:
for sub in tqdm.tqdm(p.imap(render_sokoban,
zip(pairss,
[image_mode]*threads))):
pre_images_sub = sub[0]
suc_images_sub = sub[1]
pre_images.extend(pre_images_sub)
suc_images.extend(suc_images_sub)
pre_images = np.array(pre_images)
suc_images = np.array(suc_images)
print(pre_images.shape)
print("max",pre_images.max(),"min",pre_images.min())
# shuffling
random_indices = np.arange(len(pre_images))
nr.shuffle(random_indices)
pre_images = pre_images[random_indices]
suc_images = suc_images[random_indices]
if not objects:
# whole image
np.savez_compressed(path,pres=pre_images,sucs=suc_images)
return
# image
tile = 16
B, H, W, C = pre_images.shape
pre_images = image_to_tiled_objects(pre_images, tile)
suc_images = image_to_tiled_objects(suc_images, tile)
bboxes = tiled_bboxes(B, H//tile, W//tile, tile)
print(pre_images.shape,bboxes.shape)
# prune the unreachable regions
if not egocentric:
pre_images = pre_images[:,relevant]
suc_images = suc_images[:,relevant]
bboxes = bboxes[:,relevant]
print(pre_images.shape,bboxes.shape)
# note: bbox can be reused for pres and sucs
picsize = [H,W,C]
np.savez_compressed(path,pres=pre_images,sucs=suc_images,bboxes=bboxes,picsize=picsize,max_g=max_g)