WorkBuddy
c89bfe2af7
P2-1 no_data 门控依赖文案子串(脆弱): - context_builder 新增 SliceResult(text/has_data/n_records), 5 个切片函数改为显式声明 has_data - base._slice_has_data() 优先取结构化结果,str 返回仍走文案回退 (兼容既有测试 mock 与自定义切片) - build_context 的 has_stats/has_injuries 直接取切片声明 P2-2 agent_weights 无校验即落库: - validation 新增 AgentWeightsSchema / validate_agent_weights: 未知专家名丢弃、越界值钳制、总和非 1 时归一化 - orchestrator 落库前对 agent_weights 做校验 P2-3 1x2 与比分不一致被静默修正: - 仍以比分修正,但补 logger.warning 暴露 LLM 自相矛盾 P2-5/P2-6 prompt 缓存不可刷新 + 缓存键不含模板内容: - 新增 clear_prompt_cache() 供改模板后显式失效 - 缓存键纳入模板内容 hash,模板一改缓存自动失效 P2-7 ingest/backtest/settle 接口无鉴权: - 新增 require_admin_key 依赖(X-API-Key), ADMIN_API_KEY 未设置时放行并告警(不破坏本地开发) - 挂到 3 个 ingest 接口 + backtest + eval/settle P2-8 前端请求竞态 + 未使用游标分页: - Matches.tsx 用递增 seq 丢弃过期响应,避免旧筛选结果覆盖新筛选 - 接入后端已有的 cursor 分页 + 「加载更多」按钮 附带: .env.example 补齐 LLM_TIMEOUT / 分档模型 / ADMIN_API_KEY; tests 新增 10 个用例覆盖 P2-1/2/3。
79 lines
2.7 KiB
Python
79 lines
2.7 KiB
Python
"""回测路由。"""
|
|
from __future__ import annotations
|
|
|
|
import logging
|
|
|
|
from fastapi import APIRouter, Depends, HTTPException
|
|
from pydantic import BaseModel, Field
|
|
|
|
from src.api.deps import require_admin_key
|
|
from src.llm.backtest import run_backtest
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
router = APIRouter(prefix="/api/v1", tags=["backtest"])
|
|
|
|
|
|
class BacktestRequest(BaseModel):
|
|
league_id: int | None = Field(None, description="联赛 ID")
|
|
date_from: str | None = Field(None, description="起始日期 YYYY-MM-DD")
|
|
date_to: str | None = Field(None, description="结束日期 YYYY-MM-DD")
|
|
mode: str = Field("single", description="预测模式: single(快) / multi(多 agent)")
|
|
limit: int = Field(20, ge=1, le=200, description="最大回测场数")
|
|
model: str | None = Field(None, description="指定模型 (空=默认)")
|
|
|
|
|
|
@router.post("/backtest", dependencies=[Depends(require_admin_key)])
|
|
async def backtest(req: BacktestRequest):
|
|
"""对历史比赛运行回测。
|
|
|
|
该接口会对每场已完赛比赛各发起一次 LLM 预测,成本高 —— 因此需要
|
|
`X-API-Key` 鉴权(见审查报告 P2-7)。
|
|
|
|
对每场已完赛比赛:
|
|
1. 用比赛之前的数据构建上下文 (防未来信息泄漏)
|
|
2. 调 LLM 预测
|
|
3. 用实际比分回填
|
|
4. 统计准确率 / RMSE / 校准度
|
|
"""
|
|
try:
|
|
summary = await run_backtest(
|
|
league_id=req.league_id,
|
|
date_from=req.date_from,
|
|
date_to=req.date_to,
|
|
mode=req.mode,
|
|
limit=req.limit,
|
|
model=req.model,
|
|
)
|
|
except Exception as e:
|
|
logger.exception("backtest failed")
|
|
raise HTTPException(500, "回测执行失败,请查看服务器日志")
|
|
|
|
return {
|
|
"summary": {
|
|
"total": summary.total,
|
|
"scored": summary.scored,
|
|
"accuracy_1x2": summary.accuracy_1x2,
|
|
"avg_score_rmse": summary.avg_score_rmse,
|
|
"avg_subjective_confidence": summary.avg_subjective_confidence,
|
|
"calibration": summary.calibration,
|
|
},
|
|
"results": [
|
|
{
|
|
"match_id": r.match_id,
|
|
"league_code": r.league_code,
|
|
"home_team": r.home_team,
|
|
"away_team": r.away_team,
|
|
"match_date": r.match_date,
|
|
"actual_score": f"{r.actual_home}-{r.actual_away}",
|
|
"actual_1x2": r.actual_1x2,
|
|
"pred_home": r.pred_home,
|
|
"pred_away": r.pred_away,
|
|
"pred_1x2": r.pred_1x2,
|
|
"subjective_confidence": r.subjective_confidence,
|
|
"correct_1x2": r.correct_1x2,
|
|
}
|
|
for r in summary.results
|
|
],
|
|
}
|