Spaces:
Sleeping
Sleeping
File size: 3,599 Bytes
45c1565 | 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 | """
task2_medium.py
───────────────
Task 2 (Medium) — Anomaly Timing Detection
The agent observes a 7-day behavioral time series and must decide
on which day to raise its first alert (action >= 1).
Grading (continuous 0.0–1.0):
Alert on exact anomaly day → 1.00
Alert within ±1 day → 0.75
Alert within ±2 days → 0.50
Alert within ±3 days → 0.25
Alert outside anomaly window → 0.00
No alert raised in 7 days → 0.00
"""
from __future__ import annotations
from typing import List, Dict, Any, Optional
import sys, os
sys.path.insert(0, os.path.dirname(os.path.dirname(__file__)))
from env.cognitive_env import CogTraceEnv
from env.patient_simulator import PatientConfig, make_medium_patient
TASK_ID = "task2_medium"
TASK_NAME = "Anomaly Timing Detection"
DIFFICULTY = "medium"
DESCRIPTION = (
"Observe a 7-day stream of patient behavioral signals. "
"Raise an alert (action 1–3) on the day you detect an anomaly. "
"Scored on how close your first alert is to the true anomaly onset."
)
def _timing_score(alert_day: Optional[int], anomaly_day: int) -> float:
if alert_day is None:
return 0.0
diff = abs(alert_day - anomaly_day)
if diff == 0: return 1.00
if diff == 1: return 0.75
if diff == 2: return 0.50
if diff == 3: return 0.25
return 0.00
def run_episode(env: CogTraceEnv, agent_fn) -> Dict[str, Any]:
"""
Run a 7-step episode and find the first alerting step.
agent_fn : callable(observation_dict, step: int) -> int
Returns action 0–3 for each step.
"""
obs = env.reset()
first_alert_day: Optional[int] = None
trajectory = []
for step in range(env.config.episode_length):
action = int(agent_fn(obs.model_dump(), step))
action = max(0, min(3, action))
next_obs, reward, done, info = env.step(action)
trajectory.append({
"step": step,
"action": action,
"anomaly_active": info.anomaly_active,
"reward": reward.value,
})
if action > 0 and first_alert_day is None:
first_alert_day = step
obs = next_obs
if done:
break
anomaly_day = env._sim.anomaly_day
score = _timing_score(first_alert_day, anomaly_day)
return {
"first_alert_day": first_alert_day,
"anomaly_day": anomaly_day,
"score": score,
"trajectory": trajectory,
}
def grade(trajectory: List[Dict[str, Any]]) -> float:
"""
Grade a list of episode result dicts.
Each dict must contain:
- "first_alert_day" : int or None
- "anomaly_day" : int
Returns mean timing score in [0.0, 1.0].
"""
if not trajectory:
return 0.0
scores = [_timing_score(r["first_alert_day"], r["anomaly_day"]) for r in trajectory]
return round(sum(scores) / len(scores), 4)
def build_env(seed: int = 0) -> CogTraceEnv:
sim = make_medium_patient(seed)
return CogTraceEnv(config=sim.config)
if __name__ == "__main__":
import random
results = []
for seed in range(20):
env = build_env(seed)
result = run_episode(
env,
agent_fn=lambda obs, step: random.choices([0, 0, 0, 1], k=1)[0]
)
results.append(result)
final = grade(results)
print(f"[Task 2] Random baseline score: {final:.4f}")
print("Expected ~0.15–0.30 (random rarely hits the right day)")
|