Spaces:
Runtime error
Runtime error
File size: 17,547 Bytes
f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 23bb099 f9dc9d5 | 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 | from __future__ import annotations
import json
import statistics
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple
from uuid import uuid4
from utils.patchad_filter import PatchADFilter
# 可选导入主时序模型(如果不存在则使用 None)
try:
import sys
base_dir = Path(__file__).parent.parent
if str(base_dir) not in sys.path:
sys.path.insert(0, str(base_dir))
from wearable_anomaly_detector import WearableAnomalyDetector
except ImportError:
WearableAnomalyDetector = None
@dataclass
class ValidationResult:
passed: bool
errors: List[str]
warnings: List[str]
def to_dict(self) -> Dict[str, Any]:
return {
"passed": self.passed,
"errors": self.errors,
"warnings": self.warnings,
}
class CaseBuilder:
"""
根据规范化 payload 生成 case JSON + Markdown 输入描述。
支持配置主时序模型确认(自动/手工模式):
- enabled=true, mode=auto: 自动调用主时序模型确认风险
- enabled=false: 只用 PatchAD 分数,不调用主时序模型
支持两种模式:
- 模式A:平台有PatchAD,直接传入所有数据(window_data + history_windows)
- 模式B:平台没有PatchAD,通过 event_id 从 PrecheckServer 缓存获取 window_data
"""
def __init__(
self,
config_path: Optional[Path] = None,
detector: Optional[Any] = None,
precheck_server: Optional[Any] = None
) -> None:
"""
初始化 CaseBuilder
参数:
config_path: 配置文件路径,如果为None则使用默认配置
detector: 可选的主时序模型检测器实例(如果外部已加载)
precheck_server: 可选的 PrecheckServer 实例,用于模式B从缓存获取 window_data
"""
self.patchad = PatchADFilter()
self.detector = detector # 外部传入的检测器(可选)
self.precheck_server = precheck_server # PrecheckServer 实例(用于模式B)
self.config = self._load_config(config_path)
# 如果配置为自动且未传入 detector,则自动加载
if self.config.get("main_model_confirmation", {}).get("enabled", False):
if self.detector is None and WearableAnomalyDetector is not None:
model_dir = self.config["main_model_confirmation"].get("model_dir")
if model_dir:
# 转换为绝对路径(相对于 hf_release 目录)
base_dir = Path(__file__).parent.parent
model_path = base_dir / model_dir if not Path(model_dir).is_absolute() else Path(model_dir)
device = self.config["main_model_confirmation"].get("device", "cpu")
threshold = self.config["main_model_confirmation"].get("threshold")
try:
self.detector = WearableAnomalyDetector(
model_dir=model_path,
device=device,
threshold=threshold
)
print(f"✅ 已自动加载主时序模型: {model_path}")
except Exception as e:
print(f"⚠️ 自动加载主时序模型失败: {e},将只使用 PatchAD 分数")
self.detector = None
def _load_config(self, config_path: Optional[Path]) -> Dict[str, Any]:
"""加载配置文件"""
if config_path is None:
config_path = Path(__file__).parent.parent / "configs" / "case_builder_config.json"
if config_path.exists():
try:
with open(config_path, 'r', encoding='utf-8') as f:
return json.load(f)
except Exception as e:
print(f"⚠️ 加载 CaseBuilder 配置失败: {e},使用默认配置")
# 返回默认配置(向后兼容)
return {
"main_model_confirmation": {
"enabled": False,
"mode": "auto",
"model_dir": "checkpoints/phase2/exp_factor_balanced",
"device": "cpu",
"threshold": None,
"min_duration_days": 3
}
}
def validate_payload(self, payload: Dict[str, Any]) -> ValidationResult:
errors: List[str] = []
warnings: List[str] = []
if not payload.get("user_id"):
errors.append("缺少 user_id")
# 模式B:通过 event_id 从缓存获取 window_data
event_id = payload.get("event_id")
window = payload.get("window_data")
if event_id and self.precheck_server and not window:
# 模式B:从 PrecheckServer 缓存获取 window_data(不删除,因为可能还需要)
if self.precheck_server.has_pending(event_id):
# 使用 pop_event 获取并删除(因为已经使用过了)
pending = self.precheck_server.pop_event(event_id)
window = pending.get("window_data")
payload["window_data"] = window
if pending.get("user_id") and not payload.get("user_id"):
payload["user_id"] = pending.get("user_id")
# 补充 metadata 中的 patchad 信息
if not payload.get("metadata", {}).get("patchad_score"):
metadata = payload.get("metadata", {})
if "patchad_score" not in metadata:
metadata["patchad_score"] = pending.get("metadata", {}).get("patchad_score", 0.0)
if "threshold" not in metadata:
metadata["threshold"] = pending.get("metadata", {}).get("threshold", 0.0)
payload["metadata"] = metadata
# 模式A:直接传入 window_data
if not window or len(window) < 12:
errors.append("window_data 不足 12 条(模式A需直接传入,模式B需通过 event_id 从缓存获取)")
profile = payload.get("user_profile")
if not profile:
errors.append("缺少 user_profile")
history = payload.get("history_windows", [])
if not history:
errors.append("缺少 history_windows(平台需自己合成历史窗口数据)")
metadata = payload.get("metadata", {})
if not metadata.get("detector"):
warnings.append("metadata.detector 未提供,将标记为 unknown")
return ValidationResult(passed=not errors, errors=errors, warnings=warnings)
@staticmethod
def _infer_date(window: List[Dict[str, Any]]) -> str:
for point in window:
ts = point.get("timestamp")
if ts:
return ts[:10]
return datetime.now().strftime("%Y-%m-%d")
def _normalize_history(self, history_windows: List[Any]) -> List[Dict[str, Any]]:
normalized = []
for entry in history_windows:
if isinstance(entry, dict) and "window" in entry:
window = entry["window"]
date = entry.get("date") or self._infer_date(window)
else:
window = entry
date = self._infer_date(window)
normalized.append({"date": date, "window": window})
return normalized
@staticmethod
def _avg_feature(window: List[Dict[str, Any]], feature_name: str) -> float:
values = []
for point in window:
features = point.get("features", {})
value = features.get(feature_name)
if isinstance(value, (int, float)):
values.append(float(value))
return statistics.fmean(values) if values else 0.0
@staticmethod
def _extract_steps(window: List[Dict[str, Any]]) -> float:
steps = []
for point in window:
features = point.get("features", {})
for key in ("steps", "step_count", "average_steps"):
if key in features and isinstance(features[key], (int, float)):
steps.append(float(features[key]))
break
return sum(steps)
def _build_daily_results(self, normalized_history: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
daily_results = []
for item in normalized_history:
window = item["window"]
score = self.patchad.score_window(window)
daily_results.append(
{
"date": item["date"],
"hrv_rmssd": round(self._avg_feature(window, "hrv_rmssd"), 2),
"hr": round(self._avg_feature(window, "hr"), 2),
"anomaly_score": round(score, 4),
}
)
return daily_results
def _compute_baseline(self, normalized_history: List[Dict[str, Any]]) -> Dict[str, Any]:
hrv_values = []
for item in normalized_history:
for point in item["window"]:
val = point.get("features", {}).get("baseline_hrv_mean")
if isinstance(val, (int, float)):
hrv_values.append(float(val))
personal_mean = statistics.fmean(hrv_values) if hrv_values else 0.0
personal_std = statistics.pstdev(hrv_values) if len(hrv_values) > 1 else 0.0
return {
"personal_mean": round(personal_mean, 2),
"personal_std": round(personal_std, 2),
"group_mean": round(personal_mean * 0.9, 2) if personal_mean else 0.0,
"record_count": len(hrv_values),
"baseline_type": "personal",
"baseline_reliability": "high" if len(hrv_values) >= 30 else "medium",
}
@staticmethod
def _related_indicators(normalized_history: List[Dict[str, Any]]) -> Dict[str, Any]:
steps = [CaseBuilder._extract_steps(item["window"]) for item in normalized_history]
avg_steps = statistics.fmean(steps) if steps else 0.0
activity_level = "中等" if avg_steps >= 3000 else ("低" if avg_steps < 500 else "未知")
return {
"activity_level": {
"level": activity_level,
"avg_steps": round(avg_steps, 1),
"trend": "increasing" if len(steps) > 1 and steps[-1] > steps[0] else "stable",
},
"sleep_quality": {"available": False, "quality": "数据不可用"},
"stress_indicators": {"level": "unknown"},
}
def _anomaly_pattern(self, daily_results: List[Dict[str, Any]], metadata: Dict[str, Any]) -> Dict[str, Any]:
scores = [item["anomaly_score"] for item in daily_results]
if not scores:
scores = [metadata.get("patchad_score", 0.0)]
trend = "worsening"
if len(scores) >= 2:
trend = "improving" if scores[-1] < scores[0] else ("stable" if abs(scores[-1] - scores[0]) < 0.02 else "worsening")
return {
"type": metadata.get("anomaly_type", "continuous_anomaly"),
"duration_days": len(daily_results),
"trend": trend,
"min_score": min(scores),
"max_score": max(scores),
"avg_score": round(statistics.fmean(scores), 4),
"threshold": metadata.get("threshold", self.patchad.threshold),
}
def _format_llm_input(self, case: Dict[str, Any]) -> str:
lines = [
"# 健康异常检测分析报告",
"",
"## 异常概览",
f"**异常类型**:{case['anomaly_pattern']['type']} ",
f"**持续时长**:{case['anomaly_pattern']['duration_days']}天 ",
f"**异常趋势**:{case['anomaly_pattern']['trend']} ",
"",
"## 异常评分分析",
f"- **异常分数范围**:{case['anomaly_pattern']['min_score']:.4f} - {case['anomaly_pattern']['max_score']:.4f}",
f"- **平均异常分数**:{case['anomaly_pattern']['avg_score']:.4f}",
f"- **检测阈值**:{case['anomaly_pattern']['threshold']:.4f}",
"",
"## 当前生理状态评估",
"| 指标 | 当前值 | 基线值 | 偏离基线 |",
"|------|--------|--------|----------|",
]
baseline = case["baseline_info"]
current_hrv = case["daily_results"][-1]["hrv_rmssd"] if case["daily_results"] else 0.0
deviation_pct = 0.0
if baseline["personal_mean"]:
deviation_pct = (current_hrv - baseline["personal_mean"]) / baseline["personal_mean"] * 100
lines.append(f"| HRV RMSSD | {current_hrv:.2f} ms | {baseline['personal_mean']:.2f} ms | {deviation_pct:+.1f}% |")
lines.append("")
lines.append("## 历史趋势分析")
lines.append("| 日期 | HRV (ms) | 心率 (bpm) | 异常分数 |")
lines.append("|------|----------|------------|----------|")
for record in case["daily_results"]:
lines.append(
f"| {record['date']} | {record['hrv_rmssd']:.2f} | {record['hr']:.2f} | {record['anomaly_score']:.4f} |"
)
lines.append("")
lines.append("## 相关健康指标分析")
rel = case["related_indicators"]["activity_level"]
lines.append(f"- 活动水平:{rel['level']}(平均步数={rel['avg_steps']},趋势={rel['trend']})")
lines.append("- 睡眠质量:数据不可用")
lines.append("- 压力指标:暂无显著异常")
lines.append("")
lines.append("## 用户背景信息")
profile = case["user_profile"]
lines.append(f"- 年龄:{profile.get('estimated_age')}岁({profile.get('age_group')})")
lines.append(f"- 性别:{profile.get('sex')}")
lines.append(f"- 运动习惯:{profile.get('exercise')}")
lines.append(f"- 咖啡:{profile.get('coffee')} / 饮酒:{profile.get('drinking')}")
lines.append(f"- 生物节律:MEQ={profile.get('MEQ')} ({profile.get('MEQ_type')})")
return "\n".join(lines)
def build_case(self, payload: Dict[str, Any]) -> Dict[str, Any]:
validation = self.validate_payload(payload)
if not validation.passed:
raise ValueError("; ".join(validation.errors))
normalized_history = self._normalize_history(payload["history_windows"])
daily_results = self._build_daily_results(normalized_history)
baseline_info = self._compute_baseline(normalized_history)
related_indicators = self._related_indicators(normalized_history)
anomaly_pattern = self._anomaly_pattern(daily_results, payload.get("metadata", {}))
case = {
"case_id": payload.get("event_id") or f"case_{uuid4().hex[:8]}",
"user_id": payload["user_id"],
"generated_at": datetime.now(timezone.utc).isoformat(),
"anomaly_pattern": anomaly_pattern,
"baseline_info": baseline_info,
"related_indicators": related_indicators,
"daily_results": daily_results,
"user_profile": payload["user_profile"],
"metadata": payload.get("metadata", {}),
}
llm_input = self._format_llm_input(case)
# 【新增】主时序模型确认(如果配置启用)
risk_confirmed = None
main_model_result = None
if self.config.get("main_model_confirmation", {}).get("enabled", False):
if self.detector:
try:
# 使用主时序模型对历史窗口进行确认
# 将数据组织成按天的格式:[[day1_data], [day2_data], ...]
daily_data_points = []
for item in normalized_history:
daily_data_points.append(item["window"])
if daily_data_points:
pattern_result = self.detector.detect_pattern(
daily_data_points,
days=len(normalized_history),
min_duration_days=self.config["main_model_confirmation"].get("min_duration_days", 3)
)
# 判断是否确认有风险
risk_confirmed = pattern_result.get('anomaly_pattern', {}).get('has_pattern', False)
main_model_result = pattern_result
except Exception as e:
print(f"⚠️ 主时序模型确认失败: {e},将只使用 PatchAD 分数")
risk_confirmed = None
main_model_result = None
return {
"case": case,
"llm_input": llm_input,
"validation": validation.to_dict(),
"risk_confirmed": risk_confirmed, # 【新增】主时序模型确认结果
"main_model_result": main_model_result, # 【新增】主时序模型完整结果
}
def save_case_to_file(case_bundle: Dict[str, Any], output_dir: Path) -> Path:
output_dir.mkdir(parents=True, exist_ok=True)
case_id = case_bundle["case"]["case_id"]
case_path = output_dir / f"{case_id}.json"
with open(case_path, "w", encoding="utf-8") as f:
json.dump(case_bundle, f, ensure_ascii=False, indent=2)
return case_path
|