Spaces:
Build error
Build error
File size: 13,590 Bytes
9753ee2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 | """
Task Definitions and Deterministic Graders
===========================================
Provides three progressively difficult real-world disruption tasks and
task-specific grading functions that return normalized scores in [0.0, 1.0].
"""
from typing import Dict, List
from environment import EnvState, PassengerStatus, PriorityTier
EASY_TASK = {
"task_id": "easy_minor_disruption",
"difficulty": "easy",
"objective": "Rebook all passengers on same-airline flights while preserving premium service and minimizing spend.",
"max_budget": 3000,
"max_steps": 40,
"passengers": [
{
"id": "P1",
"name": "Alice Johnson",
"priority_tier": "Platinum",
"original_flight": "FL-100",
"cabin_class": "Business",
"connection_deadline_hrs": None,
},
{
"id": "P2",
"name": "Bob Smith",
"priority_tier": "Gold",
"original_flight": "FL-100",
"cabin_class": "Economy",
"connection_deadline_hrs": None,
},
{
"id": "P3",
"name": "Carol Davis",
"priority_tier": "Standard",
"original_flight": "FL-100",
"cabin_class": "Economy",
"connection_deadline_hrs": None,
},
],
"flights": [
{
"id": "FL-102",
"destination": "New York",
"departure_hrs": 3.0,
"economy_seats": 5,
"business_seats": 2,
"is_partner": False,
},
{
"id": "FL-104",
"destination": "New York",
"departure_hrs": 6.0,
"economy_seats": 10,
"business_seats": 3,
"is_partner": False,
},
{
"id": "FL-201",
"destination": "New York",
"departure_hrs": 4.0,
"economy_seats": 3,
"business_seats": 1,
"is_partner": True,
},
],
}
MEDIUM_TASK = {
"task_id": "medium_connection_crisis",
"difficulty": "medium",
"objective": "Prioritize high-tier passengers with tight deadlines under constrained seats and budget.",
"max_budget": 5000,
"max_steps": 60,
"passengers": [
{
"id": "P1",
"name": "David Lee",
"priority_tier": "Platinum",
"original_flight": "FL-300",
"cabin_class": "Business",
"connection_deadline_hrs": 4.0,
},
{
"id": "P2",
"name": "Emma Wilson",
"priority_tier": "Gold",
"original_flight": "FL-300",
"cabin_class": "Economy",
"connection_deadline_hrs": 2.5,
},
{
"id": "P3",
"name": "Frank Brown",
"priority_tier": "Silver",
"original_flight": "FL-300",
"cabin_class": "Economy",
"connection_deadline_hrs": None,
},
{
"id": "P4",
"name": "Grace Kim",
"priority_tier": "Standard",
"original_flight": "FL-300",
"cabin_class": "Business",
"connection_deadline_hrs": 5.0,
},
{
"id": "P5",
"name": "Henry Park",
"priority_tier": "Gold",
"original_flight": "FL-300",
"cabin_class": "Economy",
"connection_deadline_hrs": 3.0,
},
],
"flights": [
{
"id": "FL-302",
"destination": "Chicago",
"departure_hrs": 2.0,
"economy_seats": 2,
"business_seats": 1,
"is_partner": False,
},
{
"id": "FL-304",
"destination": "Chicago",
"departure_hrs": 5.0,
"economy_seats": 4,
"business_seats": 0,
"is_partner": False,
},
{
"id": "FL-401",
"destination": "Chicago",
"departure_hrs": 3.5,
"economy_seats": 2,
"business_seats": 1,
"is_partner": True,
},
],
}
HARD_TASK = {
"task_id": "hard_multi_wave_disruption",
"difficulty": "hard",
"objective": "Handle mixed loyalty tiers, scarce seats, and multiple urgent connections while staying under budget.",
"max_budget": 7000,
"max_steps": 90,
"passengers": [
{
"id": "P1",
"name": "Iris Patel",
"priority_tier": "Platinum",
"original_flight": "FL-500",
"cabin_class": "Business",
"connection_deadline_hrs": 2.5,
},
{
"id": "P2",
"name": "Jack Rivera",
"priority_tier": "Gold",
"original_flight": "FL-500",
"cabin_class": "Economy",
"connection_deadline_hrs": 2.0,
},
{
"id": "P3",
"name": "Karen Novak",
"priority_tier": "Gold",
"original_flight": "FL-500",
"cabin_class": "Business",
"connection_deadline_hrs": 4.0,
},
{
"id": "P4",
"name": "Liam Chen",
"priority_tier": "Silver",
"original_flight": "FL-500",
"cabin_class": "Economy",
"connection_deadline_hrs": 3.0,
},
{
"id": "P5",
"name": "Maya Brooks",
"priority_tier": "Standard",
"original_flight": "FL-500",
"cabin_class": "Economy",
"connection_deadline_hrs": None,
},
{
"id": "P6",
"name": "Noah Singh",
"priority_tier": "Platinum",
"original_flight": "FL-500",
"cabin_class": "Business",
"connection_deadline_hrs": 3.5,
},
{
"id": "P7",
"name": "Olivia Green",
"priority_tier": "Silver",
"original_flight": "FL-500",
"cabin_class": "Economy",
"connection_deadline_hrs": 5.0,
},
{
"id": "P8",
"name": "Peter Hall",
"priority_tier": "Standard",
"original_flight": "FL-500",
"cabin_class": "Economy",
"connection_deadline_hrs": 2.8,
},
],
"flights": [
{
"id": "FL-502",
"destination": "San Francisco",
"departure_hrs": 1.8,
"economy_seats": 2,
"business_seats": 1,
"is_partner": False,
},
{
"id": "FL-504",
"destination": "San Francisco",
"departure_hrs": 3.0,
"economy_seats": 2,
"business_seats": 1,
"is_partner": False,
},
{
"id": "FL-506",
"destination": "San Francisco",
"departure_hrs": 5.5,
"economy_seats": 3,
"business_seats": 0,
"is_partner": False,
},
{
"id": "FL-701",
"destination": "San Francisco",
"departure_hrs": 2.2,
"economy_seats": 2,
"business_seats": 1,
"is_partner": True,
},
{
"id": "FL-703",
"destination": "San Francisco",
"departure_hrs": 4.4,
"economy_seats": 2,
"business_seats": 1,
"is_partner": True,
},
],
}
TASKS = {
"easy": EASY_TASK,
"medium": MEDIUM_TASK,
"hard": HARD_TASK,
}
_OUTCOME_SCORES = {
PassengerStatus.REBOOKED: 1.00,
PassengerStatus.PARTNER_REBOOKED: 0.85,
PassengerStatus.DOWNGRADED: 0.65,
PassengerStatus.HOTEL_BOOKED: 0.40,
PassengerStatus.NO_SOLUTION: 0.00,
PassengerStatus.PENDING: 0.00,
}
_TIER_WEIGHTS = {
PriorityTier.PLATINUM: 4,
PriorityTier.GOLD: 3,
PriorityTier.SILVER: 2,
PriorityTier.STANDARD: 1,
}
_GRADING_PROFILES = {
"easy": {
"quality": 0.45,
"coverage": 0.20,
"connection": 0.10,
"budget": 0.15,
"policy": 0.10,
},
"medium": {
"quality": 0.38,
"coverage": 0.17,
"connection": 0.22,
"budget": 0.13,
"policy": 0.10,
},
"hard": {
"quality": 0.30,
"coverage": 0.15,
"connection": 0.30,
"budget": 0.15,
"policy": 0.10,
},
}
def _clamp(value: float) -> float:
return max(0.01, min(0.99, value))
def _resolve_tier_weight(tier: PriorityTier) -> int:
if isinstance(tier, str):
tier = PriorityTier(tier)
return _TIER_WEIGHTS.get(tier, 1)
def _resolve_outcome_score(status: PassengerStatus) -> float:
if isinstance(status, str):
status = PassengerStatus(status)
return _OUTCOME_SCORES.get(status, 0.0)
def _connection_score(state: EnvState) -> float:
deadline_passengers = [p for p in state.passengers if p.connection_deadline_hrs is not None]
if not deadline_passengers:
return 0.99
weighted_hits = 0.0
weighted_total = 0.0
flights_by_id = {f.id: f for f in state.flights}
for passenger in deadline_passengers:
weight = _resolve_tier_weight(passenger.priority_tier)
weighted_total += weight
if passenger.assigned_flight is None:
continue
flight = flights_by_id.get(passenger.assigned_flight)
if flight is None:
continue
if flight.departure_hrs <= passenger.connection_deadline_hrs:
weighted_hits += weight
else:
weighted_hits += weight * 0.2
if weighted_total <= 0:
return 0.01
return _clamp(weighted_hits / weighted_total)
def _coverage_score(state: EnvState) -> float:
if not state.passengers:
return 0.01
resolved = sum(1 for p in state.passengers if p.status != PassengerStatus.PENDING)
return _clamp(resolved / len(state.passengers))
def _quality_score(state: EnvState) -> float:
weighted_sum = 0.0
weighted_total = 0.0
for passenger in state.passengers:
weight = _resolve_tier_weight(passenger.priority_tier)
weighted_total += weight
weighted_sum += weight * _resolve_outcome_score(passenger.status)
if weighted_total <= 0:
return 0.01
return _clamp(weighted_sum / weighted_total)
def _budget_score(state: EnvState, max_budget: float) -> float:
if max_budget <= 0:
return 0.99
return _clamp(1.0 - (state.budget_spent / max_budget))
def _policy_score(state: EnvState) -> float:
invalid_actions = max(getattr(state, "invalid_actions", 0), 0)
invalid_penalty = min(invalid_actions * 0.03, 0.3)
order: Dict[str, int] = {}
step = 0
for event in state.actions_taken:
if not event.get("success", False):
continue
action = event.get("action", {})
passenger_id = action.get("passenger_id")
if passenger_id and passenger_id not in order:
order[passenger_id] = step
step += 1
inversion_pairs = 0
total_pairs = 0
passengers = list(state.passengers)
for i in range(len(passengers)):
for j in range(i + 1, len(passengers)):
p_i = passengers[i]
p_j = passengers[j]
w_i = _resolve_tier_weight(p_i.priority_tier)
w_j = _resolve_tier_weight(p_j.priority_tier)
if w_i == w_j:
continue
if p_i.id not in order or p_j.id not in order:
continue
total_pairs += 1
if w_i > w_j and order[p_i.id] > order[p_j.id]:
inversion_pairs += 1
if w_j > w_i and order[p_j.id] > order[p_i.id]:
inversion_pairs += 1
inversion_penalty = (inversion_pairs / total_pairs) if total_pairs > 0 else 0.0
return _clamp(1.0 - invalid_penalty - inversion_penalty)
def _grade_with_profile(state: EnvState, max_budget: float, profile_name: str) -> float:
profile = _GRADING_PROFILES[profile_name]
quality = _quality_score(state)
coverage = _coverage_score(state)
connection = _connection_score(state)
budget = _budget_score(state, max_budget)
policy = _policy_score(state)
final = (
profile["quality"] * quality
+ profile["coverage"] * coverage
+ profile["connection"] * connection
+ profile["budget"] * budget
+ profile["policy"] * policy
)
return _clamp(final)
def grade_easy_episode(state: EnvState, max_budget: float) -> float:
return _grade_with_profile(state, max_budget, "easy")
def grade_medium_episode(state: EnvState, max_budget: float) -> float:
return _grade_with_profile(state, max_budget, "medium")
def grade_hard_episode(state: EnvState, max_budget: float) -> float:
return _grade_with_profile(state, max_budget, "hard")
TASK_GRADERS = {
"easy": grade_easy_episode,
"medium": grade_medium_episode,
"hard": grade_hard_episode,
}
def grade_task(task_key: str, state: EnvState, max_budget: float) -> float:
grader = TASK_GRADERS[task_key]
score = grader(state, max_budget)
# Enforce strict (0, 1) bounds required by the validator
return max(0.01, min(0.99, float(score)))
def grade_episode(state: EnvState, max_budget: float) -> float:
"""Backward-compatible default grader, mapped to medium difficulty."""
score = grade_medium_episode(state, max_budget)
return max(0.01, min(0.99, float(score)))
|