File size: 24,315 Bytes
66f749a | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 | """
Simulation orchestrator - manages large-scale agent simulations.
Uses async patterns inspired by AgentSociety:
- QwenLLM actor pool for distributed LLM calls (via HuggingFace Space Ollama API)
- asyncio.gather() for concurrent agent processing
- Agents are plain objects, not Ray actors
"""
import logging
import asyncio
import json
import os
from typing import List, Dict, Any, Optional
from simulation.ray_cluster import init_ray_cluster, shutdown_ray
from simulation.agents.social_agent import SocialAgent
from simulation.utils.profile_generator import ProfileGenerator
from simulation.llm_client import QwenLLM, shutdown_llm_pool
logger = logging.getLogger(__name__)
def _clean_env(name: str, default: str = "") -> str:
"""Read env var and trim surrounding whitespace/newlines."""
return (os.getenv(name, default) or default).strip()
class SimulationOrchestrator:
"""
Orchestrate large-scale agent simulations.
Manages:
- LLM actor pool (Ray-backed, Qwen via Ollama API)
- Agent spawning (plain objects)
- Ad content distribution
- Result collection and analysis
- Risk detection
"""
def __init__(
self,
experiment_id: str,
num_agents: int = 1000,
mqtt_host: str = "localhost",
mqtt_port: int = 1883,
mqtt_transport: str = "tcp",
mqtt_path: Optional[str] = None,
chroma_host: str = "localhost",
chroma_port: int = 8000,
chroma_ssl: bool = False,
):
self.experiment_id = experiment_id
self.num_agents = num_agents
self.chroma_host = chroma_host
self.chroma_port = chroma_port
self.chroma_ssl = chroma_ssl
self.agents: List[SocialAgent] = []
self.profiles = []
self.social_network = {}
self.event_logs = []
self.llm_pool: Optional[QwenLLM] = None
async def run(
self,
ad_content: str,
demographic_filter: Optional[Dict[str, Any]] = None,
simulation_days: int = 5,
redis_client=None,
custom_agent_profiles: Optional[List[Dict]] = None,
use_custom_agents_only: bool = False,
) -> Dict[str, Any]:
"""
Run full simulation (async).
Args:
ad_content: VLM-generated description of ad
demographic_filter: Target demographics
simulation_days: Days to simulate
redis_client: Redis client for progress updates
Returns:
Simulation results with engagement score, sentiment, risk flags
"""
logger.info(
f"Starting simulation {self.experiment_id} "
f"with {self.num_agents} agents"
)
try:
# Initialize Ray cluster
init_ray_cluster(num_cpus=None)
# Create LLM actor pool (Qwen via Ollama API)
num_actors = min(4, max(1, self.num_agents // 3))
self.llm_pool = QwenLLM(num_actors=num_actors)
logger.info(f"LLM actor pool created with {num_actors} actors")
# Profiles and spawning agents
self._update_progress(redis_client, 5, 0, 0)
self._spawn_agents(custom_agent_profiles, use_custom_agents_only, demographic_filter)
logger.info(f"Spawned {len(self.agents)} agents")
# Have all agents perceive the ad
self._update_progress(redis_client, 20, 1, len(self.agents))
final_states = await self._run_simulation(
ad_content, simulation_days, redis_client
)
# Analyze results
results = self._analyze_results(final_states)
# Cleanup
self._cleanup()
logger.info(
f"Simulation complete. "
f"Engagement score: {results['engagement_score']:.1f}"
)
return results
except Exception as e:
logger.error(f"Simulation failed: {e}")
self._cleanup()
raise
def _spawn_agents(
self,
custom_agent_profiles: Optional[List[Dict]] = None,
use_custom_agents_only: bool = False,
demographic_filter: Optional[Dict[str, Any]] = None
):
"""Spawn all agents as plain Python objects and configure network"""
self.agents = []
if use_custom_agents_only:
# Only use the explicitly provided custom profiles — no AI padding
if custom_agent_profiles:
self.profiles = list(custom_agent_profiles)
else:
logger.warning(
"use_custom_agents_only=True but no custom profiles were provided. "
"Simulation will run with 0 agents."
)
self.profiles = []
elif custom_agent_profiles:
self.profiles = list(custom_agent_profiles)
remaining = self.num_agents - len(self.profiles)
if remaining > 0:
self.profiles.extend(ProfileGenerator.generate_profiles(remaining, demographic_filter))
self.profiles = self.profiles[:self.num_agents]
else:
self.profiles = ProfileGenerator.generate_profiles(self.num_agents, demographic_filter)
self.social_network = ProfileGenerator.generate_social_network(
self.profiles, avg_friends=8
)
# Try to create a shared memory store (optional)
memory_store = self._create_memory_store()
for profile in self.profiles:
agent_id = profile["agent_id"]
friends = self.social_network.get(agent_id, [])
agent = SocialAgent(
agent_id=agent_id,
profile=profile,
experiment_id=self.experiment_id,
llm_pool=self.llm_pool,
friends=friends,
memory_store=memory_store,
)
self.agents.append(agent)
def _create_memory_store(self):
"""Try to create a shared ChromaDB memory store (optional)"""
try:
from simulation.agents.agent_memory import AgentMemoryStore
return AgentMemoryStore(
chroma_host=self.chroma_host,
chroma_port=self.chroma_port,
collection_name=f"exp_{self.experiment_id}",
ssl=self.chroma_ssl,
)
except Exception as e:
logger.info(
f"ChromaDB not available, running without memory: {e}"
)
return None
async def _run_simulation(
self,
ad_content: str,
simulation_days: int,
redis_client=None,
) -> List[Dict[str, Any]]:
"""
Run the actual simulation using asyncio.gather().
Each agent perceives the ad via async calls to the LLM actor pool.
Rate limiting is applied between batches.
"""
logger.info("Broadcasting ad content to all agents...")
# Qwen HF Space has no per-minute rate limits, but responses are slow
# (~30-60s per request on free CPU). Using batch_size=3 with 2s delay.
batch_size = 3
batch_delay = 2
all_states = []
total_batches = (len(self.agents) + batch_size - 1) // batch_size
logger.info(
f"Processing {len(self.agents)} agents in {total_batches} "
f"batches (Qwen via Ollama API)"
)
for i in range(0, len(self.agents), batch_size):
batch = self.agents[i : i + batch_size]
batch_num = (i // batch_size) + 1
# Check for cancellation
if redis_client:
try:
if redis_client.get(f"sim:{self.experiment_id}:cancel"):
logger.warning(f"Simulation {self.experiment_id} was cancelled by user.")
raise Exception("Cancelled by user")
except Exception as e:
if str(e) == "Cancelled by user":
raise e
logger.debug(f"Redis check failed: {e}")
logger.info(
f"Batch {batch_num}/{total_batches}: "
f"Processing {len(batch)} agents..."
)
# Use asyncio.gather for concurrent processing within batch
tasks = [agent.perceive_ad(ad_content) for agent in batch]
batch_states = await asyncio.gather(*tasks, return_exceptions=True)
# Collect results, handle any exceptions
for j, state in enumerate(batch_states):
if isinstance(state, Exception):
logger.error(
f"Agent {batch[j].agent_id} failed: {state}"
)
all_states.append(batch[j].get_state())
else:
all_states.append(state)
logger.info(
f" ✓ {state.get('agent_id','?')}: "
f"{state.get('emotion','?')} / {state.get('opinion_on_ad','?')} "
f"— {state.get('reasoning','')[:80]}"
)
# Update progress
progress = 20 + int(
(i + len(batch)) / len(self.agents) * 60
)
agents_per_day = max(
1, len(self.agents) // max(1, simulation_days)
)
current_day = min(
(i // agents_per_day) + 1, simulation_days
)
self._update_progress(
redis_client, progress, current_day, len(all_states)
)
logger.info(
f"Processed {len(all_states)}/{len(self.agents)} agents"
)
# Rate limiting delay between batches
if i + batch_size < len(self.agents):
logger.info(f"Rate limit pause: {batch_delay}s...")
await asyncio.sleep(batch_delay)
agents_by_id = {a.agent_id: a for a in self.agents}
# Simulate social influence over days
for day in range(2, simulation_days + 1):
if redis_client:
try:
if redis_client.get(f"sim:{self.experiment_id}:cancel"):
logger.warning(f"Simulation {self.experiment_id} was cancelled by user.")
raise Exception("Cancelled by user")
except Exception as e:
if str(e) == "Cancelled by user":
raise e
logger.debug(f"Redis check failed: {e}")
logger.info(f"Day {day} Social Loop - Messaging Phase")
for agent in self.agents:
if agent.has_seen_ad and agent.opinion_on_ad is not None:
message = await agent.generate_social_message()
for friend_id in agent.friends:
friend = agents_by_id.get(friend_id)
if friend:
friend.receive_peer_message(agent.agent_id, agent.opinion_on_ad, message)
logger.info(f"Day {day} Social Loop - Deliberation Phase")
deliberation_tasks = [agent.social_deliberation(ad_content) for agent in self.agents]
await asyncio.gather(*deliberation_tasks, return_exceptions=True)
if redis_client:
try:
redis_client.setex(
f"sim:{self.experiment_id}:status",
300,
json.dumps({
"progress": int((day / simulation_days) * 100),
"current_day": day,
"active_agents": len(self.agents)
})
)
except Exception as e:
logger.debug(f"Failed to update progress: {e}")
# Get final states (direct calls — agents are local objects)
self._update_progress(
redis_client, 95, simulation_days, len(all_states)
)
final_states = [agent.get_state() for agent in self.agents]
# Get event logs
self.event_logs = []
for agent in self.agents:
self.event_logs.extend(agent.get_event_log())
return final_states
def _analyze_results(
self, final_states: List[Dict[str, Any]]
) -> Dict[str, Any]:
"""Calculate engagement score, sentiment breakdown, and risk flags"""
opinions = [
s.get("opinion") for s in final_states if s.get("opinion")
]
sentiment_counts = {
"positive": sum(1 for o in opinions if o == "POSITIVE"),
"neutral": sum(1 for o in opinions if o == "NEUTRAL"),
"negative": sum(1 for o in opinions if o == "NEGATIVE"),
}
total = len(opinions) or 1
# Engagement score: high if strong reactions (positive OR negative)
strong_reactions = (
sentiment_counts["positive"] + sentiment_counts["negative"]
)
engagement_score = (strong_reactions / total) * 100
# Detect controversies
risk_flags = self._detect_controversies(final_states)
# Prepare agent logs for storage
agent_logs = self.event_logs[:1000]
# Extract opinion trajectory for top 50 agents
opinion_trajectory = {}
for state in final_states[:50]:
agent_id = state.get("agent_id")
opinion_history = state.get("opinion_history")
if agent_id and opinion_history is not None:
opinion_trajectory[agent_id] = opinion_history
# Build map data (lightweight for map rendering)
map_data = []
for state in final_states:
profile = state.get("profile", {})
agent_id = state.get("agent_id", "")
map_data.append({
"agent_id": agent_id,
"coordinates": profile.get("coordinates", [0, 0]),
"opinion": state.get("opinion", "NEUTRAL"),
"friends": self.social_network.get(agent_id, []),
})
# Build enriched agent states (for detail popups)
agent_states = []
for state in final_states:
profile = state.get("profile", {})
agent_states.append({
"agent_id": state.get("agent_id", ""),
"coordinates": profile.get("coordinates", [0, 0]),
"opinion": state.get("opinion", "NEUTRAL"),
"emotion": state.get("emotion", "neutral"),
"emotion_intensity": state.get("emotion_intensity", 0),
"reasoning": state.get("reasoning", ""),
"friends": self.social_network.get(state.get("agent_id", ""), []),
"profile": {
"name": profile.get("name"),
"age": profile.get("age"),
"gender": profile.get("gender"),
"location": profile.get("location"),
"occupation": profile.get("occupation"),
"education": profile.get("education"),
"income_level": profile.get("income_level"),
"religion": profile.get("religion"),
"ethnicity": profile.get("ethnicity"),
"social_media_usage": profile.get("social_media_usage"),
"political_leaning": profile.get("political_leaning"),
"personality_traits": profile.get("personality_traits", []),
"values": profile.get("values", []),
},
})
return {
"engagement_score": round(engagement_score, 2),
"sentiment_breakdown": sentiment_counts,
"total_agents": len(final_states),
"responding_agents": len(opinions),
"risk_flags": risk_flags,
"agent_logs": agent_logs,
"map_data": map_data,
"agent_states": agent_states,
"opinion_trajectory": opinion_trajectory,
}
def _detect_controversies(
self, final_states: List[Dict[str, Any]]
) -> List[Dict[str, Any]]:
"""Identify controversial reactions by demographic segment"""
flags = []
# Add an 'overall' bucket to catch widespread backlash regardless of demographic
groups = {"overall": {"All Agents": []}, "age": {}, "gender": {}, "location": {}, "values": {}}
for state in final_states:
profile = state.get("profile", {})
opinion = state.get("opinion")
if not opinion:
continue
# Overall
groups["overall"]["All Agents"].append(state)
# Age groups
age = profile.get("age", 0)
age_bracket = f"{(age // 10) * 10}-{(age // 10) * 10 + 9}"
groups["age"].setdefault(age_bracket, []).append(state)
# Gender
gender = profile.get("gender", "Unknown")
groups["gender"].setdefault(gender, []).append(state)
# Location
location = profile.get("location", "Unknown")
groups["location"].setdefault(location, []).append(state)
# Values
for value in profile.get("values", []):
groups["values"].setdefault(value, []).append(state)
total_responding = len(groups["overall"]["All Agents"])
min_group_size = 5 if total_responding >= 20 else 1
overall_states = groups["overall"]["All Agents"]
overall_negative_rate = sum(1 for s in overall_states if s.get("opinion") == "NEGATIVE") / max(1, total_responding)
seen_reasonings = set()
for group_type, group_data in groups.items():
for group_name, states in group_data.items():
if len(states) < min_group_size and group_type != "overall":
continue
negative_count = sum(1 for s in states if s.get("opinion") == "NEGATIVE")
total = len(states)
negative_rate = negative_count / max(1, total)
# Skip if not generally negative
if negative_rate < 0.5:
continue
# If this is a subset demographic, only flag it distinctively if it is notably worse than the general population.
# If the general population already hates it (e.g. overall = 90%), we don't need to report every single trait individually.
if group_type != "overall" and negative_rate < (overall_negative_rate + 0.15):
continue
if negative_rate > 0.8:
severity = "CRITICAL"
elif negative_rate > 0.7:
severity = "HIGH"
elif negative_rate > 0.6:
severity = "MEDIUM"
else:
severity = "LOW"
sample_reactions = []
for s in states:
if s.get("opinion") == "NEGATIVE":
reasoning = s.get("reasoning", "")[:120]
if reasoning and reasoning not in seen_reasonings:
sample_reactions.append({
"agent_id": s.get("agent_id"),
"reasoning": reasoning
})
seen_reasonings.add(reasoning)
if len(sample_reactions) >= 3:
break
# Fallback if no uniquely new reasonings exist
if not sample_reactions:
sample_reactions = [
{"agent_id": s.get("agent_id"), "reasoning": s.get("reasoning", "")[:120]}
for s in states if s.get("opinion") == "NEGATIVE"
][:3]
flags.append({
"flag_type": f"{group_type.upper()}_BACKLASH",
"severity": severity,
"description": f"{int(negative_rate * 100)}% of {group_type}={group_name} reacted negatively",
"affected_demographics": {group_type: group_name},
"sample_agent_reactions": sample_reactions,
})
severity_order = {"CRITICAL": 0, "HIGH": 1, "MEDIUM": 2, "LOW": 3}
flags.sort(key=lambda x: severity_order.get(x["severity"], 4))
return flags[:10]
def _update_progress(
self, redis_client, progress: int, current_day: int,
active_agents: int
):
"""Update progress in Redis for frontend polling"""
if redis_client:
try:
status = {
"progress": progress,
"current_day": current_day,
"active_agents": active_agents,
}
redis_client.setex(
f"sim:{self.experiment_id}:status",
60,
json.dumps(status),
)
except Exception as e:
logger.debug(f"Failed to update progress: {e}")
def _cleanup(self):
"""Clean up resources"""
self.agents = []
shutdown_llm_pool()
shutdown_ray()
async def run_simulation_async(
experiment_id: str,
ad_content: str,
demographic_filter: Optional[Dict[str, Any]] = None,
num_agents: int = 10,
simulation_days: int = 5,
redis_client=None,
custom_agent_profiles: Optional[List[Dict]] = None,
use_custom_agents_only: bool = False,
) -> Dict[str, Any]:
"""
Async convenience function to run a simulation.
"""
mqtt_host = _clean_env("MQTT_BROKER_HOST", "localhost")
mqtt_port = int(_clean_env("MQTT_BROKER_PORT", "1883"))
mqtt_transport = _clean_env("MQTT_TRANSPORT", "tcp")
mqtt_path = _clean_env("MQTT_PATH", "") or None
chroma_host = _clean_env("CHROMA_HOST", "localhost")
chroma_port = int(_clean_env("CHROMA_PORT", "8000"))
chroma_ssl = _clean_env("CHROMA_SSL", "False").lower() in ("true", "1", "yes")
orchestrator = SimulationOrchestrator(
experiment_id=experiment_id,
num_agents=num_agents,
mqtt_host=mqtt_host,
mqtt_port=mqtt_port,
mqtt_transport=mqtt_transport,
mqtt_path=mqtt_path,
chroma_host=chroma_host,
chroma_port=chroma_port,
chroma_ssl=chroma_ssl,
)
return await orchestrator.run(
ad_content=ad_content,
demographic_filter=demographic_filter,
simulation_days=simulation_days,
redis_client=redis_client,
custom_agent_profiles=custom_agent_profiles,
use_custom_agents_only=use_custom_agents_only,
)
def run_simulation(
experiment_id: str,
ad_content: str,
demographic_filter: Optional[Dict[str, Any]] = None,
num_agents: int = 10,
simulation_days: int = 5,
redis_client=None,
custom_agent_profiles: Optional[List[Dict]] = None,
use_custom_agents_only: bool = False,
) -> Dict[str, Any]:
"""
Synchronous convenience function to run a simulation.
Called from Celery task or test scripts.
"""
return asyncio.run(
run_simulation_async(
experiment_id=experiment_id,
ad_content=ad_content,
demographic_filter=demographic_filter,
num_agents=num_agents,
simulation_days=simulation_days,
redis_client=redis_client,
custom_agent_profiles=custom_agent_profiles,
use_custom_agents_only=use_custom_agents_only,
)
)
|