scripts/factor_priors.py

#!/usr/bin/env python3
from __future__ import annotations
"""
factor_priors.py — PRISM-VQ A1: 13 经典截面先验因子计算模块

从 yfinance OHLCV、yfinance info、akshare 财务数据 计算 13 个经典截面因子。
每个因子输出统一格式：{"status": "ok"/"error", "value": <float or None>, "detail": "..."}

支持 --use-cache 模式：从 priors_cache/ 目录读取预下载的 OHLCV、info、financial 数据。

防泄露核心规则：
  - compute_all_priors() 使用 df_ohlcv 切片确保只用到截止 date 的数据
  - 财务数据用最近一个财报期的值
  - yfinance info 是快照数据，只用在推理日期已知的信息
"""

import logging
import os
import re
import threading
import time
from functools import lru_cache
from pathlib import Path

import numpy as np
import pandas as pd
import yfinance as yf

logger = logging.getLogger(__name__)

# ── CSI 300 benchmark symbol ──
CACHE_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "priors_cache")

# ── CSI 300 benchmark symbol (fallback 用) ──
BENCHMARK_SYMBOL = "000300.SS"

# ── Benchmark data cache (CSI 300 fallback) ──
_benchmark_close: pd.Series | None = None
_benchmark_ts: float = 0
BENCHMARK_CACHE_TTL = 3600  # 1 hour

# ── 全市场等权收益率缓存 ──
_market_returns_cache: pd.Series | None = None
_market_returns_ts: float = 0
MARKET_RETURNS_CACHE_TTL = 3600


def _to_yf(symbol: str) -> str:
    """Convert internal symbol (sh600000) to yfinance format (600000.SS)."""
    if symbol.endswith(".SS") or symbol.endswith(".SZ"):
        return symbol
    if len(symbol) > 2:
        prefix, code = symbol[:2], symbol[2:]
        if prefix == "sh":
            return f"{code}.SS"
        elif prefix == "sz":
            return f"{code}.SZ"
    return symbol


# ===================================================================
# 缓存读取函数（新增）
# ===================================================================


def _get_cache_dir(symbol: str) -> str:
    """获取单个股票的缓存目录路径。"""
    return os.path.join(CACHE_DIR, symbol)


def load_cached_ohlcv(symbol: str) -> pd.DataFrame | None:
    """从缓存读取 OHLCV 数据。

    Args:
        symbol: 股票代码 (e.g., "sh600000")

    Returns:
        DataFrame with columns [date, open, high, low, close, volume], 或 None
    """
    path = os.path.join(_get_cache_dir(symbol), "ohlcv.parquet")
    if not os.path.exists(path):
        return None
    try:
        df = pd.read_parquet(path)
        if df.empty:
            return None
        df["date"] = pd.to_datetime(df["date"])
        return df
    except Exception as e:
        logger.warning(f"读取 OHLCV 缓存失败 [{symbol}]: {e}")
        return None


def load_cached_info(symbol: str) -> dict[str, float | None] | None:
    """从缓存读取 yfinance info 字段。

    Returns:
        {field: value} 字典，或 None
    """
    path = os.path.join(_get_cache_dir(symbol), "info.parquet")
    if not os.path.exists(path):
        return None
    try:
        df = pd.read_parquet(path)
        if df.empty:
            return None
        record = df.iloc[0].to_dict()
        # 过滤掉 symbol 列，只保留数值字段
        return {k: v for k, v in record.items() if k != "symbol"}
    except Exception as e:
        logger.warning(f"读取 Info 缓存失败 [{symbol}]: {e}")
        return None


def load_cached_financial(symbol: str) -> dict[str, float | None] | None:
    """从缓存读取 akshare 财务数据。

    Returns:
        {roe, gross_margin, asset_growth} 字典，或 None
    """
    path = os.path.join(_get_cache_dir(symbol), "financial.parquet")
    if not os.path.exists(path):
        return None
    try:
        df = pd.read_parquet(path)
        if df.empty:
            return None
        record = df.iloc[0].to_dict()
        return {
            "roe": record.get("roe"),
            "gross_margin": record.get("gross_margin"),
            "asset_growth": record.get("asset_growth"),
        }
    except Exception as e:
        logger.warning(f"读取 Financial 缓存失败 [{symbol}]: {e}")
        return None


def load_cached_market_returns() -> pd.Series | None:
    """从缓存读取全市场等权日收益率序列。

    Returns:
        pd.Series with datetime index and market_return values, or None
    """
    global _market_returns_cache, _market_returns_ts
    now = time.time()
    if _market_returns_cache is not None and (now - _market_returns_ts) < MARKET_RETURNS_CACHE_TTL:
        return _market_returns_cache

    path = os.path.join(CACHE_DIR, "market_returns.parquet")
    if not os.path.exists(path):
        return None
    try:
        df = pd.read_parquet(path)
        if df.empty or "market_return" not in df.columns:
            return None
        series = df["market_return"].sort_index()
        series.index = pd.to_datetime(series.index)
        _market_returns_cache = series
        _market_returns_ts = now
        return series
    except Exception as e:
        logger.warning(f"读取市场收益率缓存失败: {e}")
        return None


# ===================================================================
# 基准收益率（保留 CSI 300 作为后备）
# ===================================================================


def _get_benchmark_returns(end_date: pd.Timestamp, max_periods: int = 504) -> np.ndarray:
    """Get CSI 300 daily returns up to end_date (防数据泄露), cached for TTL.
    Uses yfinance first, falls back to akshare for A-share indices.
    """
    global _benchmark_close, _benchmark_ts
    now = time.time()
    if _benchmark_close is None or (now - _benchmark_ts) > BENCHMARK_CACHE_TTL:
        # Try yfinance first
        try:
            bm = yf.download(BENCHMARK_SYMBOL, period="1y", progress=False, auto_adjust=True)
            if bm.empty or len(bm) < 100:
                raise ValueError("insufficient data from yfinance")
            _benchmark_close = bm["Close"].dropna()
            _benchmark_close.index = pd.to_datetime(_benchmark_close.index).tz_localize(None)
            _benchmark_ts = now
        except Exception:
            # Fallback to akshare
            try:
                import akshare as ak
                raw = ak.stock_zh_index_daily(symbol="sh000300")
                if raw is not None and not raw.empty:
                    raw["date"] = pd.to_datetime(raw["date"])
                    raw = raw.sort_values("date")
                    _benchmark_close = raw.set_index("date")["close"]
                    _benchmark_ts = now
                else:
                    logger.warning("Benchmark: akshare returned empty")
                    return np.array([], dtype=np.float64)
            except Exception as e2:
                logger.warning(f"Benchmark all sources failed: {e2}")
                return np.array([], dtype=np.float64)

    end_ts = pd.Timestamp(end_date).tz_localize(None) if hasattr(pd.Timestamp(end_date), "tz") else pd.Timestamp(end_date)
    mask = _benchmark_close.index <= end_ts
    bm_slice = _benchmark_close[mask]
    if len(bm_slice) < 2:
        return np.array([], dtype=np.float64)
    ret = bm_slice.pct_change().dropna().values
    # Flatten in case of 2D (e.g. MultiIndex columns from yfinance)
    if ret.ndim == 2 and ret.shape[1] == 1:
        ret = ret.flatten()
    return ret[-max_periods:] if len(ret) > max_periods else ret


# ===================================================================
# Individual factor computation helpers
# ===================================================================

def _compute_momentum_12m(df_slice: pd.DataFrame) -> dict:
    """过去12个月收益减去最近1个月收益.

    经典动量因子 (Jegadeesh & Titman 1993):
    momentum = r_12m_ago_to_1m_ago  (skip most recent month)
    """
    try:
        close = df_slice["close"].dropna().values
        if len(close) < 21:
            return {"status": "error", "value": None, "detail": "insufficient data (<21 rows)"}
        n_12m = min(252, len(close))
        n_1m = min(21, len(close))
        val = close[-1] / close[-n_12m] - 1 - (close[-1] / close[-n_1m] - 1)
        return {
            "status": "ok",
            "value": float(val),
            "detail": f"n_12m={n_12m}, n_1m={n_1m}",
        }
    except Exception as e:
        return {"status": "error", "value": None, "detail": str(e)}


def _compute_reversal_1m(df_slice: pd.DataFrame) -> dict:
    """过去1个月收益 (短期反转因子)."""
    try:
        close = df_slice["close"].dropna().values
        if len(close) < 21:
            return {"status": "error", "value": None, "detail": "insufficient data (<21 rows)"}
        val = close[-1] / close[-21] - 1
        return {"status": "ok", "value": float(val), "detail": ""}
    except Exception as e:
        return {"status": "error", "value": None, "detail": str(e)}


def _compute_volatility_idio(df_slice: pd.DataFrame, market_returns: np.ndarray | None = None) -> dict:
    """残差波动率: 对市场回归后的残差标准差.

    Args:
        df_slice: OHLCV 切片
        market_returns: 可选，全市场等权收益率序列 (numpy array)
    """
    try:
        close = df_slice["close"].dropna()
        if len(close) < 30:
            return {"status": "error", "value": None, "detail": "insufficient data (<30 rows)"}
        stock_ret = close.pct_change().dropna().values

        end_date = pd.Timestamp(df_slice["date"].iloc[-1])
        if market_returns is not None and len(market_returns) > 20:
            bm_ret = market_returns
        else:
            bm_ret = _get_benchmark_returns(end_date, 504)
        if len(bm_ret) < 20:
            return {"status": "error", "value": None, "detail": f"insufficient benchmark data ({len(bm_ret)})"}

        # Align lengths to the shorter series
        min_len = min(len(stock_ret), len(bm_ret))
        stock_ret = stock_ret[-min_len:]
        bm_ret = bm_ret[-min_len:]

        # Remove any NaN/inf values
        mask = np.isfinite(stock_ret) & np.isfinite(bm_ret)
        stock_ret = stock_ret[mask]
        bm_ret = bm_ret[mask]

        if len(stock_ret) < 20:
            return {"status": "error", "value": None, "detail": f"too few valid returns ({len(stock_ret)})"}

        X = np.column_stack([np.ones(len(stock_ret), dtype=np.float64), bm_ret])
        y = stock_ret
        beta = np.linalg.lstsq(X, y, rcond=None)[0]
        residuals = y - X @ beta
        ivol = float(np.std(residuals))
        return {"status": "ok", "value": ivol, "detail": f"beta={beta[1]:.4f}, alpha={beta[0]:.6f}, n={len(stock_ret)}"}
    except Exception as e:
        return {"status": "error", "value": None, "detail": str(e)}


def _compute_beta_60d(df_slice: pd.DataFrame, market_returns: np.ndarray | None = None) -> dict:
    """60 日市场 beta。

    优先使用全市场等权收益率作为基准 (从缓存读取)，
    如果不可用则回退到 CSI 300。

    beta = Cov(R_i, R_m) / Var(R_m)

    Args:
        df_slice: OHLCV 切片
        market_returns: 可选，全市场等权收益率序列 (numpy array)
    """
    try:
        close = df_slice["close"]
        if len(close) < 61:
            return {"status": "error", "value": None, "detail": "insufficient data (<61 rows)"}

        stock_ret = close.pct_change().dropna().values[-60:]

        # 优先使用全市场等权收益率
        if market_returns is not None and len(market_returns) >= 20:
            bm_ret = market_returns[-60:] if len(market_returns) >= 60 else market_returns
        else:
            end_date = pd.Timestamp(df_slice["date"].iloc[-1])
            bm_ret = _get_benchmark_returns(end_date, 60)

        if len(bm_ret) < 20:
            return {"status": "error", "value": None, "detail": f"insufficient benchmark data ({len(bm_ret)})"}

        min_len = min(len(stock_ret), len(bm_ret))
        stock_ret = stock_ret[-min_len:]
        bm_ret = bm_ret[-min_len:]

        if min_len < 10:
            return {"status": "error", "value": None, "detail": "insufficient data after alignment"}

        cov = np.cov(stock_ret, bm_ret)
        beta = cov[0, 1] / cov[1, 1] if cov[1, 1] > 1e-12 else 0.0
        return {"status": "ok", "value": float(beta), "detail": f"n={min_len}"}
    except Exception as e:
        return {"status": "error", "value": None, "detail": str(e)}


def _compute_turnover_avg(df_slice: pd.DataFrame, shares_outstanding: int | None) -> dict:
    """20 日平均换手率 = 平均日成交量 / 流通股数."""
    try:
        volume = df_slice["volume"].values
        if len(volume) < 20:
            return {"status": "error", "value": None, "detail": "insufficient data (<20 rows)"}
        avg_vol = float(np.mean(volume[-20:]))
        if shares_outstanding is None or shares_outstanding <= 0:
            return {"status": "error", "value": None, "detail": "shares_outstanding not available"}
        turnover = avg_vol / shares_outstanding
        return {
            "status": "ok",
            "value": float(turnover),
            "detail": f"avg_vol={avg_vol:.0f}, shares={shares_outstanding}",
        }
    except Exception as e:
        return {"status": "error", "value": None, "detail": str(e)}


# ===================================================================
# yfinance info helpers (with LRU cache)
# ===================================================================

@lru_cache(maxsize=4096)
def _get_info_field_cached(symbol: str, field: str) -> tuple[str, float | None, str]:
    """Cached wrapper for single info field fetch.
    Checks local priors_cache/yfinance_info.parquet first, falls back to yfinance.info.
    """
    # Try cache first
    try:
        cache_path = Path(__file__).parent / "priors_cache" / "yfinance_info.parquet"
        if cache_path.exists():
            # Use LRU cache on the parquet read itself
            if not hasattr(_get_info_field_cached, "_cache_df"):
                _get_info_field_cached._cache_df = pd.read_parquet(cache_path)
            cache_df = _get_info_field_cached._cache_df
            # Combined cache uses internal format (sh600000); symbol may be yfinance format (600000.SS)
            lookup_keys = [symbol]
            if symbol.endswith(".SS"):
                lookup_keys.append("sh" + symbol[:-3])
            elif symbol.endswith(".SZ"):
                lookup_keys.append("sz" + symbol[:-3])
            row = cache_df[cache_df["symbol"].isin(lookup_keys)]
            if len(row) > 0 and field in row.columns:
                val = row.iloc[0][field]
                if val is not None and not (isinstance(val, float) and np.isnan(val)):
                    try:
                        return ("ok", float(val), "cached")
                    except (TypeError, ValueError):
                        pass
    except Exception:
        pass

    # Fallback to yfinance
    try:
        ticker = yf.Ticker(symbol)
        info = ticker.info
        if info and field in info and info[field] is not None:
            try:
                val = float(info[field])
                return ("ok", val, "")
            except (TypeError, ValueError):
                return ("error", None, f"field '{field}' value not numeric: {info[field]}")
        else:
            return ("error", None, f"field '{field}' not in info")
    except Exception as e:
        return ("error", None, str(e))


@lru_cache(maxsize=4096)
def _get_shares_outstanding_cached(symbol: str) -> int | None:
    """Get shares outstanding from yfinance info (cached)."""
    try:
        ticker = yf.Ticker(symbol)
        info = ticker.info
        if info:
            for field in ["sharesOutstanding", "impliedSharesOutstanding", "floatShares"]:
                val = info.get(field)
                if val is not None and val > 0:
                    return int(val)
        return None
    except Exception:
        return None


def _compute_size(yf_sym: str, cached_info: dict | None = None) -> dict:
    """log(Market Cap)."""
    if cached_info is not None:
        val = cached_info.get("marketCap")
        if val is not None and val > 0:
            return {"status": "ok", "value": float(np.log(val)), "detail": f"raw_marketCap={val:.0f} (cached)"}
    status, val, detail = _get_info_field_cached(yf_sym, "marketCap")
    if status == "ok" and val is not None and val > 0:
        return {"status": "ok", "value": float(np.log(val)), "detail": f"raw_marketCap={val:.0f}"}
    return {"status": status, "value": None, "detail": detail}


def _compute_pe(yf_sym: str, cached_info: dict | None = None) -> dict:
    """市盈率 (Price/Earnings)."""
    if cached_info is not None:
        val = cached_info.get("trailingPE")
        if val is not None:
            return {"status": "ok", "value": val, "detail": "cached"}
    status, val, detail = _get_info_field_cached(yf_sym, "trailingPE")
    return {"status": status, "value": val, "detail": detail}


def _compute_pb(yf_sym: str, cached_info: dict | None = None) -> dict:
    """市净率 (Price/Book)."""
    if cached_info is not None:
        val = cached_info.get("priceToBook")
        if val is not None:
            return {"status": "ok", "value": val, "detail": "cached"}
    status, val, detail = _get_info_field_cached(yf_sym, "priceToBook")
    return {"status": status, "value": val, "detail": detail}


def _compute_ps(yf_sym: str, cached_info: dict | None = None) -> dict:
    """市销率 (Price/Sales)."""
    if cached_info is not None:
        val = cached_info.get("priceToSalesTrailing12Months")
        if val is not None:
            return {"status": "ok", "value": val, "detail": "cached"}
    status, val, detail = _get_info_field_cached(yf_sym, "priceToSalesTrailing12Months")
    return {"status": status, "value": val, "detail": detail}


def _compute_dividend_yield(yf_sym: str, cached_info: dict | None = None) -> dict:
    """股息率."""
    if cached_info is not None:
        val = cached_info.get("dividendYield")
        if val is not None:
            return {"status": "ok", "value": val, "detail": "cached"}
        val = cached_info.get("trailingAnnualDividendYield")
        if val is not None:
            return {"status": "ok", "value": val, "detail": "cached (trailing)"}

    status, val, detail = _get_info_field_cached(yf_sym, "dividendYield")
    if status == "ok":
        return {"status": status, "value": val, "detail": detail}
    status, val, detail = _get_info_field_cached(yf_sym, "trailingAnnualDividendYield")
    return {"status": status, "value": val, "detail": detail}


def _get_shares_from_cache_or_api(yf_sym: str, cached_info: dict | None = None) -> int | None:
    """从缓存或 API 获取流通股数。"""
    if cached_info is not None:
        for field in ["sharesOutstanding", "impliedSharesOutstanding", "floatShares"]:
            val = cached_info.get(field)
            if val is not None and val > 0:
                return int(val)
    return _get_shares_outstanding_cached(yf_sym)


# ===================================================================
# akshare financial data（带超时保护）
# ===================================================================

@lru_cache(maxsize=8192)
def _compute_akshare_financial(symbol: str) -> dict:
    """从 akshare 获取最新年度财务摘要: ROE, 毛利率, 总资产增长率.
    优先读缓存（priors_cache/{symbol}/financial.parquet）。

    Returns:
        dict with status, value (dict with keys roe/gross_margin/asset_growth or None), detail
    """
    # Try cache first
    cached = load_cached_financial(symbol)
    if cached is not None:
        return {
            "status": "ok",
            "value": {"roe": cached["roe"], "gross_margin": cached["gross_margin"],
                       "asset_growth": cached["asset_growth"]},
            "detail": "from cache"
        }

    # No cache — skip akshare (can hang), return None gracefully
    return {"status": "error", "value": None, "detail": "not in cache, akshare skipped"}

    sym_num = re.sub(r"[^0-9]", "", symbol)
    if not sym_num:
        return {"status": "error", "value": None, "detail": f"cannot parse symbol: {symbol}"}

    errors = []
    result = {"roe": None, "gross_margin": None, "asset_growth": None}

    # ── Part 1: ROE and Gross Margin from stock_financial_abstract ──
    try:
        df = ak.stock_financial_abstract(symbol=sym_num)
        if df is not None and not df.empty and '指标' in df.columns:
            date_cols = [c for c in df.columns if c not in ('选项', '指标') and re.match(r'^\d{8}$', str(c))]
            if not date_cols:
                date_cols = sorted([c for c in df.columns if c not in ('选项', '指标')], reverse=True)
            else:
                date_cols = sorted(date_cols, reverse=True)

            if date_cols:
                for _, row in df.iterrows():
                    indicator = str(row['指标'])
                    val = None
                    for dc in date_cols:
                        v = row.get(dc)
                        try:
                            vf = float(v) if v is not None else None
                        except (ValueError, TypeError):
                            vf = None
                        if vf is not None and np.isfinite(vf):
                            val = vf
                            break
                    if val is None:
                        continue
                    if '净资产收益率' in indicator and 'ROE' in indicator:
                        if result['roe'] is None:
                            result['roe'] = val
                    elif indicator == '毛利率':
                        if result['gross_margin'] is None:
                            result['gross_margin'] = val

            if any(v is not None for v in result.values()):
                errors.append("stock_financial_abstract: partial OK")
            else:
                errors.append("stock_financial_abstract: all fields missing")
        else:
            errors.append("stock_financial_abstract: empty or missing '指标' column")
    except Exception as e:
        errors.append(f"stock_financial_abstract: {e}")

    # ── Part 2: Asset growth from stock_financial_debt_ths ──
    try:
        df_debt = ak.stock_financial_debt_ths(symbol=sym_num, indicator='按年度')
        if df_debt is not None and not df_debt.empty:
            ta_col = None
            for candidate in ['*资产合计', '资产合计', '资产总计']:
                if candidate in df_debt.columns:
                    ta_col = candidate
                    break
            if ta_col is None:
                for c in df_debt.columns:
                    if '资产合计' in str(c) or '资产总计' in str(c):
                        ta_col = c
                        break

            if ta_col:
                period_col = '报告期'
                if period_col in df_debt.columns:
                    def _parse_chinese_num(s):
                        if s is None:
                            return None
                        s = str(s).strip()
                        if not s:
                            return None
                        try:
                            return float(s)
                        except ValueError:
                            pass
                        if '万亿' in s:
                            try:
                                return float(s.replace('万亿', '').replace(',', '').strip()) * 1e12
                            except ValueError:
                                pass
                        if '亿' in s:
                            try:
                                return float(s.replace('亿', '').replace(',', '').strip()) * 1e8
                            except ValueError:
                                pass
                        if '万' in s:
                            try:
                                return float(s.replace('万', '').replace(',', '').strip()) * 1e4
                            except ValueError:
                                pass
                        return None

                    years = []
                    for _, row in df_debt.iterrows():
                        try:
                            y = int(str(row[period_col])[:4])
                            years.append(y)
                        except (ValueError, TypeError):
                            years.append(None)
                    df_debt = df_debt.copy()
                    df_debt['_year'] = years
                    df_debt = df_debt.dropna(subset=['_year'])
                    df_debt = df_debt.sort_values('_year', ascending=False)

                    if len(df_debt) >= 2:
                        latest_ta = _parse_chinese_num(df_debt[ta_col].iloc[0])
                        prev_ta = _parse_chinese_num(df_debt[ta_col].iloc[1])
                        if latest_ta is not None and prev_ta is not None and prev_ta > 0:
                            growth = (latest_ta / prev_ta - 1) * 100
                            result['asset_growth'] = float(growth)
                            errors.append("asset_growth computed from stock_financial_debt_ths")
            else:
                errors.append("stock_financial_debt_ths: no total assets column found")
        else:
            errors.append("stock_financial_debt_ths: empty result")
    except Exception as e:
        errors.append(f"stock_financial_debt_ths: {e}")

    if any(v is not None for v in result.values()):
        return {"status": "ok", "value": result, "detail": "; ".join(errors)}
    return {"status": "error", "value": None, "detail": "; ".join(errors) if errors else "all akshare APIs failed"}


def _timeout_akshare(symbol: str, timeout: int = 30) -> dict:
    """带超时保护的 akshare 财务数据下载。"""
    result_container = [{"status": "error", "value": None, "detail": "timeout"}]

    def runner():
        try:
            result_container[0] = _compute_akshare_financial(symbol)
        except Exception as e:
            result_container[0] = {"status": "error", "value": None, "detail": str(e)}

    t = threading.Thread(target=runner, daemon=True)
    t.start()
    t.join(timeout)
    return result_container[0]


# ===================================================================
# Main entry point
# ===================================================================

PRIOR_FACTOR_NAMES = [
    "momentum_12m",
    "reversal_1m",
    "volatility_idio",
    "beta_60d",
    "turnover_avg",
    "size",
    "pe",
    "pb",
    "ps",
    "dividend_yield",
    "roe",
    "gross_margin",
    "asset_growth",
]

PRIOR_COLUMNS = [f"prior_{n}" for n in PRIOR_FACTOR_NAMES]


def compute_all_priors(
    df_ohlcv: pd.DataFrame,
    symbol: str,
    date: pd.Timestamp,
    use_cache: bool = False,
) -> dict[str, float | None]:
    """计算单只股票在某个日期的全部 13 个先验因子值.

    使用 df_ohlcv 中截止到 date 的数据（防数据泄露）.

    Args:
        df_ohlcv: OHLCV DataFrame with columns [date, open, high, low, close, volume].
        symbol: 股票代码 (e.g., "sh600000").
        date: 计算截止日期。
        use_cache: 是否优先从 priors_cache 读取信息/财务数据。

    Returns:
        dict with prior_* keys and float values (None if computation failed).
    """
    # Ensure date column is datetime (tz-naive) and slice up to date
    if "date" in df_ohlcv.columns:
        df = df_ohlcv.copy()
        df["date"] = pd.to_datetime(df["date"]).dt.tz_localize(None)
        target = pd.Timestamp(date).tz_localize(None) if hasattr(pd.Timestamp(date), "tz") else pd.Timestamp(date)
        df_slice = df[df["date"] <= target].sort_values("date").reset_index(drop=True)
    else:
        df_slice = df_ohlcv.copy()

    if len(df_slice) < 2:
        return {f"prior_{k}": None for k in PRIOR_FACTOR_NAMES}

    # Convert symbol to yfinance format
    yf_sym = _to_yf(symbol)

    # ── 从缓存读取可选数据 ──
    cached_info: dict | None = None
    cached_financial: dict | None = None
    market_returns_series: pd.Series | None = None

    if use_cache:
        cached_info = load_cached_info(symbol)
        cached_financial = load_cached_financial(symbol)
        market_returns_series = load_cached_market_returns()

    # 将 market_returns 转换为 numpy array（对齐到日期）
    market_returns_arr: np.ndarray | None = None
    if market_returns_series is not None and not market_returns_series.empty:
        target_date = pd.Timestamp(date).tz_localize(None) if hasattr(pd.Timestamp(date), "tz") else pd.Timestamp(date)
        mr = market_returns_series[market_returns_series.index <= target_date]
        if len(mr) > 10:
            market_returns_arr = mr.values[-252:]  # 最多 1 年数据

    # ── Price-based factors ──
    momentum = _compute_momentum_12m(df_slice)
    reversal = _compute_reversal_1m(df_slice)
    vol_idio = _compute_volatility_idio(df_slice, market_returns_arr)
    beta = _compute_beta_60d(df_slice, market_returns_arr)

    # Turnover needs shares outstanding
    if use_cache and cached_info is not None:
        shares_out = _get_shares_from_cache_or_api(yf_sym, cached_info)
    else:
        shares_out = _get_shares_outstanding_cached(yf_sym)
    turnover = _compute_turnover_avg(df_slice, shares_out)

    # ── yfinance info-based factors ──
    size_val = _compute_size(yf_sym, cached_info)
    pe_val = _compute_pe(yf_sym, cached_info)
    pb_val = _compute_pb(yf_sym, cached_info)
    ps_val = _compute_ps(yf_sym, cached_info)
    dy_val = _compute_dividend_yield(yf_sym, cached_info)

    # ── akshare financial factors — 被 A1 验证排除（静态值无时间信号，拖累 IC）
    # 详见: https://github.com/cedricwyh/Project-Jinjing/issues/... （TODO）
    roe_val = {"status": "error", "value": None, "detail": "excluded in A1 — static values hurt IC"}
    gm_val = {"status": "error", "value": None, "detail": "excluded in A1"}
    ag_val = {"status": "error", "value": None, "detail": "excluded in A1"}

    # ── Compile results ──
    results = {}
    for name, result in [
        ("prior_momentum_12m", momentum),
        ("prior_reversal_1m", reversal),
        ("prior_volatility_idio", vol_idio),
        ("prior_beta_60d", beta),
        ("prior_turnover_avg", turnover),
        ("prior_size", size_val),
        ("prior_pe", pe_val),
        ("prior_pb", pb_val),
        ("prior_ps", ps_val),
        ("prior_dividend_yield", dy_val),
        ("prior_roe", roe_val),
        ("prior_gross_margin", gm_val),
        ("prior_asset_growth", ag_val),
    ]:
        if isinstance(result, dict):
            results[name] = result.get("value")
        else:
            results[name] = result

    return results


def compute_all_priors_batch(
    symbols: list[str],
    ohlcv_dict: dict[str, pd.DataFrame],
    date: pd.Timestamp,
    verbose: bool = True,
    use_cache: bool = False,
) -> pd.DataFrame:
    """批量计算多个股票的先验因子.

    Args:
        symbols: 股票代码列表
        ohlcv_dict: {symbol: OHLCV DataFrame} 字典
        date: 计算截止日期
        verbose: 是否显示进度
        use_cache: 是否优先从缓存读取

    Returns:
        DataFrame with columns [symbol] + prior_* columns
    """
    rows = []
    iterator = enumerate(symbols)
    if verbose:
        try:
            from tqdm import tqdm
            iterator = tqdm(list(enumerate(symbols)), desc="Priors")
        except ImportError:
            pass

    for idx, sym in iterator:
        ohlcv = ohlcv_dict.get(sym)
        if ohlcv is None or len(ohlcv) < 2:
            row = {"symbol": sym}
            for pcol in PRIOR_COLUMNS:
                row[pcol] = None
            rows.append(row)
            continue

        priors = compute_all_priors(ohlcv, sym, date, use_cache=use_cache)
        row = {"symbol": sym}
        row.update(priors)
        rows.append(row)

        if not verbose and (idx + 1) % 100 == 0:
            logger.info(f"  Priors: {idx + 1}/{len(symbols)}")

    return pd.DataFrame(rows)