Spaces:
Sleeping
Sleeping
File size: 15,133 Bytes
dbc6675 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 | """Plan recovery, corruption, rollout, and labeling utilities."""
from __future__ import annotations
import json
import os
import random
import re
import subprocess
import tempfile
from dataclasses import dataclass
from pathlib import Path
from typing import Callable
PREDICATE_REGEX = re.compile(r"\(([\w-]+(?: [\w-]+)*)\)")
WHITESPACE_REGEX = re.compile(r"\s+")
@dataclass(frozen=True)
class ProblemContext:
"""Parsed symbolic context for one planning problem."""
domain: str
split: str
problem: str
domain_path: Path
problem_path: Path
task: object
operators: tuple[object, ...]
operator_by_name: dict[str, object]
objects: tuple[str, ...]
goal_atoms: tuple[str, ...]
def canonical_action(action: str) -> str:
"""Normalize action strings for pyperplan and VAL interoperability."""
text = WHITESPACE_REGEX.sub(" ", str(action).strip().lower())
if not text:
return text
if not text.startswith("("):
text = f"({text})"
return text
def parse_trajectory_state(line: str) -> frozenset[str]:
"""Parse one `.traj` line into the atom-string format used by pyperplan."""
return frozenset(f"({match.strip()})" for match in PREDICATE_REGEX.findall(line.strip()))
def load_problem_context(
*,
data_dir: str | Path,
domain: str,
split: str,
problem: str,
) -> ProblemContext:
"""Parse and ground a PDDL problem."""
from pyperplan.grounding import ground
from pyperplan.pddl.parser import Parser
data_root = Path(data_dir)
domain_path = data_root / "pddl" / domain / "domain.pddl"
problem_path = data_root / "pddl" / domain / split / f"{problem}.pddl"
parser = Parser(str(domain_path), str(problem_path))
parsed_domain = parser.parse_domain()
parsed_problem = parser.parse_problem(parsed_domain)
task = ground(parsed_problem)
operators = tuple(sorted(task.operators, key=lambda op: op.name))
operator_by_name = {canonical_action(op.name): op for op in operators}
objects = tuple(_extract_objects(parsed_problem, parsed_domain))
goal_atoms = tuple(str(goal) for goal in task.goals)
return ProblemContext(
domain=domain,
split=split,
problem=problem,
domain_path=domain_path,
problem_path=problem_path,
task=task,
operators=operators,
operator_by_name=operator_by_name,
objects=objects,
goal_atoms=goal_atoms,
)
def _extract_objects(parsed_problem, parsed_domain) -> list[str]:
"""Extract sorted object and constant names from pyperplan structures."""
objects: set[str] = set()
problem_objects = getattr(parsed_problem, "objects", {})
if isinstance(problem_objects, dict):
objects.update(str(name) for name in problem_objects.keys())
else:
for obj in problem_objects:
objects.add(obj.name if hasattr(obj, "name") else str(obj))
constants = getattr(parsed_domain, "constants", {})
if isinstance(constants, dict):
objects.update(str(name) for name in constants.keys())
else:
for obj in constants:
objects.add(obj.name if hasattr(obj, "name") else str(obj))
return sorted(objects)
def load_trajectory_states(traj_path: str | Path) -> list[frozenset[str]]:
"""Load a symbolic state trajectory from disk."""
with open(traj_path, "r", encoding="utf-8") as f:
return [parse_trajectory_state(line) for line in f if line.strip()]
def recover_plan_from_states(
states: list[frozenset[str]],
operators: tuple[object, ...],
) -> list[str]:
"""Recover the unique grounded action that connects each trajectory step."""
recovered: list[str] = []
for step_idx, (current, next_state) in enumerate(zip(states, states[1:])):
matches = [
op.name
for op in operators
if op.applicable(set(current)) and frozenset(op.apply(set(current))) == next_state
]
if len(matches) != 1:
raise RuntimeError(
f"Could not recover a unique action at step {step_idx}: "
f"found {len(matches)} matches"
)
recovered.append(canonical_action(matches[0]))
return recovered
def recover_gold_plan(
*,
data_dir: str | Path,
domain: str,
split: str,
problem: str,
) -> list[str]:
"""Recover a gold action plan from the existing state trajectory."""
context = load_problem_context(
data_dir=data_dir,
domain=domain,
split=split,
problem=problem,
)
traj_path = Path(data_dir) / "states" / domain / split / f"{problem}.traj"
states = load_trajectory_states(traj_path)
return recover_plan_from_states(states, context.operators)
def rollout_plan(context: ProblemContext, plan: list[str]) -> list[frozenset[str]]:
"""
Roll out a candidate plan into a state sequence.
Inapplicable or unknown actions are represented as stalled no-op transitions.
This keeps feature extraction fixed-length without passing explicit symbolic
failure flags to the downstream classifier.
"""
states = [frozenset(context.task.initial_state)]
current = set(context.task.initial_state)
for action in plan:
op = context.operator_by_name.get(canonical_action(action))
if op is not None and op.applicable(current):
current = set(op.apply(current))
states.append(frozenset(current))
return states
def label_plan_internal(context: ProblemContext, plan: list[str]) -> tuple[bool, bool]:
"""
Label a plan with pyperplan semantics.
This is intended for smoke tests when VAL is unavailable. Paper-quality
dataset builds should use VAL.
"""
current = set(context.task.initial_state)
executable = True
for action in plan:
op = context.operator_by_name.get(canonical_action(action))
if op is None or not op.applicable(current):
executable = False
break
current = set(op.apply(current))
valid = executable and set(context.task.goals).issubset(current)
return bool(valid), bool(executable)
def validate_candidate_plan(
*,
domain_path: str | Path,
problem_path: str | Path,
plan: list[str],
val_path: str | Path,
) -> tuple[bool, bool]:
"""Label a candidate plan using VAL."""
if not plan:
return False, False
val_bin = Path(val_path)
if not val_bin.exists() or not os.access(val_bin, os.X_OK):
raise FileNotFoundError(f"VAL binary not found or not executable: {val_path}")
with tempfile.NamedTemporaryFile(mode="w", delete=False, suffix=".plan") as tmp:
for action in plan:
tmp.write(f"{canonical_action(action)}\n")
tmp_plan_path = Path(tmp.name).resolve()
try:
result = subprocess.run(
[
str(val_bin),
"-v",
str(Path(domain_path).resolve()),
str(Path(problem_path).resolve()),
str(tmp_plan_path),
],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
text=True,
check=False,
)
output = result.stdout or ""
if result.returncode != 0 and not output.strip():
raise RuntimeError(
f"VAL failed without output for {problem_path}; "
f"exit code {result.returncode}. Check VAL runtime dependencies."
)
is_valid = "Plan valid" in output
is_executable = is_valid or "Plan executed successfully" in output
return is_valid, is_executable
finally:
if tmp_plan_path.exists():
tmp_plan_path.unlink()
def corruption_types() -> list[str]:
"""Supported negative-plan corruption operators."""
return ["truncate", "delete", "swap", "replace", "insert", "repeat"]
def corrupt_plan(
gold_plan: list[str],
all_actions: list[str],
rng: random.Random,
corruption_type: str,
) -> list[str]:
"""Create one corrupted action sequence."""
plan = list(gold_plan)
if not plan:
return plan
if corruption_type == "truncate":
cut = rng.randint(0, max(0, len(plan) - 1))
return plan[:cut]
if corruption_type == "delete":
idx = rng.randrange(len(plan))
return plan[:idx] + plan[idx + 1 :]
if corruption_type == "swap":
if len(plan) == 1:
return corrupt_plan(plan, all_actions, rng, "replace")
idx = rng.randrange(len(plan) - 1)
plan[idx], plan[idx + 1] = plan[idx + 1], plan[idx]
return plan
if corruption_type == "replace":
idx = rng.randrange(len(plan))
choices = [action for action in all_actions if action != plan[idx]]
if not choices:
return plan
plan[idx] = rng.choice(choices)
return plan
if corruption_type == "insert":
idx = rng.randrange(len(plan) + 1)
return plan[:idx] + [rng.choice(all_actions)] + plan[idx:]
if corruption_type == "repeat":
start = rng.randrange(len(plan))
end = rng.randint(start + 1, min(len(plan), start + 3))
insert_at = rng.randrange(len(plan) + 1)
segment = plan[start:end]
return plan[:insert_at] + segment + plan[insert_at:]
raise ValueError(f"Unknown corruption type: {corruption_type}")
def make_candidate_id(
domain: str,
split: str,
problem: str,
variant_index: int,
corruption_type: str,
) -> str:
"""Build a stable candidate identifier."""
return f"{domain}::{split}::{problem}::{variant_index:03d}::{corruption_type}"
def generate_labeled_candidates_for_problem(
*,
context: ProblemContext,
gold_plan: list[str],
val_path: str | Path | None = None,
negative_ratio: int,
rng: random.Random,
validator: Validator | None = None,
max_attempts_per_negative: int = 25,
) -> list[dict]:
"""Create one positive and several VAL-confirmed negative candidates."""
candidates: list[dict] = []
seen_plans = {tuple(gold_plan)}
label = validator or _make_val_validator(context, val_path)
is_valid, is_executable = label(gold_plan)
candidates.append(
{
"candidate_id": make_candidate_id(
context.domain,
context.split,
context.problem,
0,
"gold",
),
"domain": context.domain,
"split": context.split,
"problem": context.problem,
"corruption_type": "gold",
"plan": gold_plan,
"plan_len": len(gold_plan),
"gold_plan_len": len(gold_plan),
"label_valid": int(is_valid),
"label_executable": int(is_executable),
}
)
action_space = [canonical_action(op.name) for op in context.operators]
type_cycle = corruption_types()
variant_index = 1
attempts = 0
max_attempts = max_attempts_per_negative * max(1, negative_ratio)
while variant_index <= negative_ratio and attempts < max_attempts:
attempts += 1
corruption_type = type_cycle[(attempts - 1) % len(type_cycle)]
candidate_plan = corrupt_plan(gold_plan, action_space, rng, corruption_type)
plan_key = tuple(candidate_plan)
if plan_key in seen_plans:
continue
seen_plans.add(plan_key)
is_valid, is_executable = label(candidate_plan)
if is_valid:
continue
candidates.append(
{
"candidate_id": make_candidate_id(
context.domain,
context.split,
context.problem,
variant_index,
corruption_type,
),
"domain": context.domain,
"split": context.split,
"problem": context.problem,
"corruption_type": corruption_type,
"plan": candidate_plan,
"plan_len": len(candidate_plan),
"gold_plan_len": len(gold_plan),
"label_valid": 0,
"label_executable": int(is_executable),
}
)
variant_index += 1
return candidates
def write_jsonl(path: str | Path, rows: list[dict]) -> None:
"""Write records as JSON lines."""
path = Path(path)
path.parent.mkdir(parents=True, exist_ok=True)
with open(path, "w", encoding="utf-8") as f:
for row in rows:
f.write(json.dumps(row) + "\n")
def read_jsonl(path: str | Path) -> list[dict]:
"""Read records from JSON lines."""
with open(path, "r", encoding="utf-8") as f:
return [json.loads(line) for line in f if line.strip()]
def discover_val_path(repo_root: str | Path, user_val_path: str | None = None) -> str | None:
"""Resolve the VAL binary used for plan labels."""
if user_val_path:
return user_val_path
root = Path(repo_root)
candidates = [
root / "VAL" / "build" / "bin" / "Validate.exe",
root / "VAL" / "build" / "bin" / "Validate",
root / "VAL" / "bin" / "Validate.exe",
root / "VAL" / "bin" / "Validate",
root / "VAL" / "VAL" / "build" / "bin" / "Validate.exe",
]
for candidate in candidates:
if candidate.exists() and os.access(candidate, os.X_OK):
return str(candidate)
return None
def iter_problem_names(
*,
data_dir: str | Path,
domain: str,
split: str,
max_problems: int | None = None,
) -> list[str]:
"""Return problem names with both trajectory and PDDL files available."""
state_dir = Path(data_dir) / "states" / domain / split
pddl_dir = Path(data_dir) / "pddl" / domain / split
if not state_dir.exists() or not pddl_dir.exists():
return []
names = []
for traj_path in sorted(state_dir.glob("*.traj")):
if (pddl_dir / f"{traj_path.stem}.pddl").exists():
names.append(traj_path.stem)
if max_problems is not None:
names = names[:max_problems]
return names
def summarize_labels(rows: list[dict]) -> dict[str, int]:
"""Small summary for CLI status messages and metadata."""
total = len(rows)
valid = sum(int(row["label_valid"]) for row in rows)
executable = sum(int(row["label_executable"]) for row in rows)
return {
"total": total,
"valid": valid,
"invalid": total - valid,
"executable": executable,
"non_executable": total - executable,
}
Validator = Callable[[list[str]], tuple[bool, bool]]
def _make_val_validator(context: ProblemContext, val_path: str | Path | None) -> Validator:
if val_path is None:
raise ValueError("val_path is required when no custom validator is provided.")
def validate(plan: list[str]) -> tuple[bool, bool]:
return validate_candidate_plan(
domain_path=context.domain_path,
problem_path=context.problem_path,
plan=plan,
val_path=val_path,
)
return validate
|