File size: 11,445 Bytes
8a685c0 350500c 8a685c0 abcf7c3 350500c abcf7c3 350500c 8a685c0 350500c 8a685c0 350500c 8a685c0 350500c abcf7c3 350500c abcf7c3 8a685c0 350500c 8a685c0 350500c abcf7c3 350500c abcf7c3 d550096 abcf7c3 d550096 abcf7c3 d550096 abcf7c3 d550096 abcf7c3 8a685c0 abcf7c3 8a685c0 abcf7c3 8a685c0 abcf7c3 8a685c0 abcf7c3 350500c abcf7c3 8a685c0 abcf7c3 350500c abcf7c3 350500c abcf7c3 8a685c0 350500c 8a685c0 abcf7c3 350500c | 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 | import requests
from openenv.core.env_server import Environment
from src.models import AdAction, AdObservation, AdState
from src.generator import AdGenerator
REGULATORY_API = "http://localhost:8001"
CRM_API = "http://localhost:8002"
AUDIT_API = "http://localhost:8003"
VALID_ACTIONS = {
"query_regulations",
"analyze_image",
"check_advertiser_history",
"request_landing_page",
"request_id_verification",
"submit_audit",
"approve",
"reject"
}
TERMINAL_ACTIONS = {"approve", "reject"}
REQUIRED_BEFORE_TERMINAL = {
"query_regulations",
"submit_audit"
}
MAX_STEPS = 8
class AdPolicyEnvironment(Environment):
def __init__(self):
super().__init__()
self.generator = AdGenerator()
self.current_ad = None
self.step_count = 0
self.total_reward = 0.0
self.actions_taken = set()
self.api_failed = False
self.api_recovered = False
self.last_failed_action = None
self.last_error = None
self.trace = []
self.signals = {
"risk_score": None,
"policy_confidence": None,
"image_flag": None,
"landing_flag": None
}
def _ensure_ad(self, task_id=None):
if self.current_ad is None:
self.current_ad = self.generator.generate_random_ad(task_id)
self.current_ad["task_id"] = task_id or "task_1_healthcare"
def state(self) -> AdState:
self._ensure_ad()
return AdState(
step_count=self.step_count,
total_reward=self.total_reward,
current_ad_id=self.current_ad.get("ad_id", "N/A")
)
def reset(self, task_id: str = None) -> AdObservation:
self.current_ad = self.generator.generate_random_ad(task_id)
self.current_ad["task_id"] = task_id or "task_1_healthcare"
self.step_count = 0
self.total_reward = 0.0
self.actions_taken = set()
self.api_failed = False
self.api_recovered = False
self.last_failed_action = None
self.last_error = None
self.trace = []
self.signals = {
"risk_score": None,
"policy_confidence": None,
"image_flag": None,
"landing_flag": None
}
return self._get_obs(f"Ad loaded for {self.current_ad['task_id']}. Begin with query_regulations.")
def step(self, action: AdAction) -> AdObservation:
self._ensure_ad()
if not action or not hasattr(action, 'action_type'):
return self._get_obs("Invalid action format.", -0.5, True)
act_type = str(action.action_type).lower()
# 1. Validate action
if act_type not in VALID_ACTIONS:
return self._get_obs(f"Invalid action: {act_type}.", -0.5, True)
# 2. Start constraint — state based
if "query_regulations" not in self.actions_taken:
if act_type != "query_regulations":
return self._get_obs("Must call query_regulations first.", -0.2, False)
self.step_count += 1
# 3. Execute action
response = self._execute_action(act_type)
# 4. Update state
if "error" in response:
self.api_failed = True
self.last_failed_action = act_type
self.last_error = response["error"]
# Notice we DO NOT add to self.actions_taken here
else:
self.actions_taken.add(act_type) # <-- ADDED HERE: Only register successful actions
if act_type == self.last_failed_action:
self.api_recovered = True
self.last_error = None
self._extract_signals(act_type, response)
# 5. Append trace
self.trace.append({
"step": self.step_count,
"action": act_type,
"result": self._summarize_response(act_type, response)
})
# 6. Compute reward
reward = -0.05 # step penalty
# 7. Handle terminal
done = False
if act_type in TERMINAL_ACTIONS:
reward += self._terminal_reward(act_type)
done = True
elif self.step_count >= MAX_STEPS:
reward -= 0.5
done = True
self.total_reward += reward
summary = self._summarize_response(act_type, response)["summary"]
return self._get_obs(summary, reward, done)
def _execute_action(self, act_type: str) -> dict:
task_id = self.current_ad.get("task_id", "")
# Deterministic failure for task_10_failure on step 1
if task_id == "task_10_failure" and self.step_count == 1:
return {"error": "service_unavailable", "retryable": True}
try:
if act_type == "query_regulations":
category = self.current_ad.get("category", "general")
resp = requests.get(f"{REGULATORY_API}/regulations/{category}", timeout=2)
return resp.json()
elif act_type == "analyze_image":
vlm_desc = self.current_ad.get("vlm_desc", "")
violation = any(kw in vlm_desc.lower() for kw in [
"violation", "banned", "prescription", "fake", "flagged",
"semaglutide", "adderall", "no rx", "no prescription"
])
return {"violation_detected": violation, "description": vlm_desc}
elif act_type == "check_advertiser_history":
advertiser_id = self.current_ad.get("advertiser_id", "adv_003")
resp = requests.get(f"{CRM_API}/advertiser/{advertiser_id}", timeout=2)
return resp.json()
elif act_type == "request_landing_page":
domain_age = self.current_ad.get("domain_age_days", 365)
risk_keywords = self.current_ad.get("landing_risk_keywords", [])
suspicious = domain_age < 30 or len(risk_keywords) > 0
return {"suspicious": suspicious, "domain_age": domain_age, "risk_keywords": risk_keywords}
elif act_type == "request_id_verification":
age_min = self.current_ad.get("targeting_data", {}).get("age_min", 18)
return {"age_min": age_min, "minor_targeted": age_min < 18}
elif act_type == "submit_audit":
payload = {
"ad_id": self.current_ad.get("ad_id", "test"),
"action_taken": "pending",
"reasoning": "audit requested"
}
resp = requests.post(f"{AUDIT_API}/log", json=payload, timeout=2)
return resp.json()
else:
return {"status": "ok"}
except Exception as e:
return {"error": f"service_unavailable", "retryable": True}
def _extract_signals(self, action: str, response: dict):
if action == "check_advertiser_history":
self.signals["risk_score"] = response.get("risk_score")
elif action == "query_regulations":
violations = response.get("violations", [])
confs = [v["confidence"] for v in violations]
self.signals["policy_confidence"] = max(confs, default=0.0)
elif action == "analyze_image":
self.signals["image_flag"] = response.get("violation_detected", False)
elif action == "request_landing_page":
self.signals["landing_flag"] = response.get("suspicious", False)
def _summarize_response(self, action: str, response: dict) -> dict:
if "error" in response:
return {"summary": "API failure — retryable", "flag": False}
if action == "check_advertiser_history":
rs = response.get("risk_score", 0.0)
return {"summary": f"risk_score={rs:.2f}", "flag": rs > 0.7}
if action == "query_regulations":
violations = response.get("violations", [])
conf = max((v["confidence"] for v in violations), default=0.0)
return {"summary": f"policy_confidence={conf:.2f}", "flag": conf > 0.7}
if action == "analyze_image":
flagged = response.get("violation_detected", False)
return {
"summary": "image_violation_detected" if flagged else "image_clean",
"flag": flagged
}
if action == "request_landing_page":
sus = response.get("suspicious", False)
return {"summary": "landing_suspicious" if sus else "landing_clean", "flag": sus}
if action == "request_id_verification":
minor = response.get("minor_targeted", False)
age = response.get("age_min", 18)
return {
"summary": f"ALERT: minor targeting age={age}" if minor else f"age_min={age} OK",
"flag": minor
}
if action == "submit_audit":
audit_id = response.get("audit_id", "LOCAL")
return {"summary": f"audit_logged id={audit_id}", "flag": False}
return {"summary": "ok", "flag": False}
def _terminal_reward(self, act_type: str) -> float:
reward = 0.0
is_violation = self.current_ad.get("ground_truth", False)
is_correct = (act_type == "reject" and is_violation) or \
(act_type == "approve" and not is_violation)
# Dominant signal
reward += 1.0 if is_correct else -1.0
# Sequence correctness
if REQUIRED_BEFORE_TERMINAL.issubset(self.actions_taken):
reward += 0.2
else:
reward -= 0.2
# API failure handling
if self.api_failed and self.api_recovered:
reward += 0.3
elif self.api_failed and not self.api_recovered:
reward -= 0.3
# Risk-aware penalty
if act_type == "approve":
rs = self.signals["risk_score"]
pc = self.signals["policy_confidence"]
if rs is not None and pc is not None:
if rs > 0.7 and pc > 0.6:
reward -= 0.5
# Ambiguity enforcement
pc = self.signals["policy_confidence"]
if pc is not None and pc < 0.6:
if "check_advertiser_history" not in self.actions_taken \
and "request_landing_page" not in self.actions_taken:
reward -= 0.4
# Landing page bonus under ambiguity
if self.signals["landing_flag"] and pc is not None and pc < 0.6:
reward += 0.2
return reward
def _get_obs(self, message, reward=0.0, done=False) -> AdObservation:
self._ensure_ad()
return AdObservation(
ad_id=str(self.current_ad.get("ad_id", "N/A")),
headline=str(self.current_ad.get("headline", "N/A")),
body_text=str(self.current_ad.get("body_text", "N/A")),
advertiser_trust_score=float(self.current_ad.get("advertiser_trust_score", 0.0)),
targeting_data=dict(self.current_ad.get("targeting_data", {})),
image_url=str(self.current_ad.get("image_url", "N/A")),
status_message=str(message),
reward=reward,
done=done,
risk_score=self.signals["risk_score"],
policy_confidence=self.signals["policy_confidence"],
image_flag=self.signals["image_flag"],
landing_flag=self.signals["landing_flag"],
last_error=self.last_error
) |