R4: run.py plan 每日选股计划首版 + score 档位步长修正(10→20)
步长 10 小于组内分全幅 19.8，档间会穿插、词典序破功；已写入的 07-29 score 表需重 build 覆盖。另修体检报表主题命中计数（去重前算）。
This commit is contained in:
zlt
parent
b813e42f52
commit
83f2b1409f
|
|
@ -254,10 +254,19 @@ probe 三张读数提醒已钉 **07-30 周四 19:00**。
|
|||
① 基座 transmission.py 传导主题黑名单（高新技术企业/小型微利企业/联营企业/无/
|
||||
报告分部——联营企业 298 名成员不触 600 闸，必须显式拦；源与目标两侧都挡）；
|
||||
② 桥 config/frontier_tracks.yml v0.1（6 confirmed + 4 candidate、61 主题映射、
|
||||
排除清单、TODO 决策点待圈）；③ 桥 tracks.py + `run.py tracks`（赛道覆盖体检 +
|
||||
排除清单、TODO 决策点待圈；当日实测：61 主题全部在池中有名，体检表"命中 x/y"
|
||||
的缺口是报表把被同赛道更早主题完全覆盖的主题误计漏，当日已修）；
|
||||
③ 桥 tracks.py + `run.py tracks`（赛道覆盖体检 +
|
||||
成员表版本化快照落 data/，即本清单"赛道覆盖体检"的正式工具）；④ 桥 akg_gate /
|
||||
akg_score 两个新因子（三档+两段式，进 build all 日更；C 闸默认关
|
||||
=ENABLE_TRACK_GATE，yml 转正后开）；⑤ requirements.txt 加 PyYAML（**桥机需
|
||||
rebuild 镜像**；不 rebuild 只影响 tracks 命令，其余照跑）。
|
||||
当日实机判收：tracks 六赛道全非空（confirmed 合计 777 只 = 池的 32.5%）、
|
||||
register 六因子成、gate 2386 行、score 1068 行（主榜 961 + 观察档 107）✅。
|
||||
**score 编码当日修正**：档位步长 10 → 20——组内分 clip±9.9 全幅 19.8，
|
||||
步长 10 会档间重叠、先档后分的词典序破功；07-29 的 score 表需重 build 覆盖。
|
||||
⑥ plan.py + `run.py plan`（R4 首版）：每日选股计划 Markdown——主榜/观察档
|
||||
榜单（名称、传导证据、热度、预期空间）+ 今日升降档 + T−1 口径声明，
|
||||
落 data/plan/，终端同步打印。
|
||||
**工作哲学（07-30 用户令）**：数据渐进式暴露是常态——不等全量抽完、不等干净
|
||||
截面、不等覆盖完备；缺锚降档不弃用，滞后交采信规则消化。
|
||||
|
|
@ -408,8 +408,9 @@ def build_score(start, end):
|
|||
主榜 = 200 + 传导档位×10 + 组内分：传导票按 log 传导分的组内中位数分成
|
||||
强(2)/弱(1)两档，无传导 0 档；组内分 = 0.6·z(−热度) + 0.4·z(upside)，
|
||||
即"还没热、还便宜"。观察档 = 100 + 0.6·z(传导) + 0.4·z(−热度)——无估值锚，
|
||||
低置信。组内分夹在 ±9.9 保证档位永不重叠；z 都在各自档内算。
|
||||
缺失处置同设计 §3.4：传导缺=0，热度缺=档内中位数。"""
|
||||
低置信。**档位步长 20、组内分夹 ±9.9**：组内分全幅 19.8 小于步长，
|
||||
先档后分的词典序才真正成立（步长 10 会档间重叠，07-30 交付当天修正）。
|
||||
z 都在各自档内算；缺失处置同设计 §3.4：传导缺=0，热度缺=档内中位数。"""
|
||||
ts = _track_set()
|
||||
out = []
|
||||
for ds in _days_of(start, end):
|
||||
|
|
@ -429,7 +430,7 @@ def build_score(start, end):
|
|||
med = lt[hit].median()
|
||||
tier[hit] = np.where(lt[hit] >= med, 2.0, 1.0)
|
||||
inner = 0.6 * (-_robust_z(P["heat"])) + 0.4 * _robust_z(P["upside"])
|
||||
score = 200.0 + tier * 10.0 + inner.clip(-9.9, 9.9)
|
||||
score = 200.0 + tier * 20.0 + inner.clip(-9.9, 9.9)
|
||||
out.append(pd.DataFrame({"trade_date": ds, "stock_code": P.index,
|
||||
"factor_value": score.values}))
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,190 @@
|
|||
"""每日选股计划（R4 首版）：把 akg_gate / akg_score 变成一份人能读的榜单。
|
||||
|
||||
数据全部来自已落库的表，不重算：
|
||||
平台因子表 t_factor_akg_score / _gate / _upside / _heat —— 当日截面
|
||||
基座只读视图 v_factor_transmission —— 传导证据（主题、源数、已动比例）
|
||||
基座 industry_pools —— 股票名称
|
||||
产出：终端打印 + data/plan/plan_<日期>.md。升降档一节对比前一交易日的档位表，
|
||||
体现"数据到达本身是信号"（首次覆盖 / 新进传导链即升档）。
|
||||
"""
|
||||
import json
|
||||
import os
|
||||
|
||||
import pandas as pd
|
||||
|
||||
import common
|
||||
import db
|
||||
|
||||
|
||||
# 分数编码（与 factors.build_score 一致）：主榜 = 200 + 传导档位×20 + 组内分，
|
||||
# 观察档 = 100 + 组内分，组内分 clip ±9.9。150 落在两带中间的空档上，用作分界。
|
||||
_MAIN_MIN = 150.0
|
||||
|
||||
|
||||
def _factor(table: str, ds: str) -> pd.Series:
|
||||
df = db.read_mysql(
|
||||
"factor",
|
||||
f"SELECT stock_code, factor_value FROM {table} WHERE trade_date = %s", (ds,))
|
||||
if df.empty:
|
||||
return pd.Series(dtype=float)
|
||||
return df.set_index("stock_code")["factor_value"].astype(float)
|
||||
|
||||
|
||||
def _latest_date(table: str, upto: str | None = None):
|
||||
if upto:
|
||||
df = db.read_mysql(
|
||||
"factor", f"SELECT MAX(trade_date) d FROM {table} "
|
||||
f"WHERE trade_date <= %s", (upto,))
|
||||
else:
|
||||
df = db.read_mysql("factor", f"SELECT MAX(trade_date) d FROM {table}")
|
||||
v = None if df.empty else df.iloc[0, 0]
|
||||
return None if v is None or pd.isna(v) else pd.Timestamp(v).date().isoformat()
|
||||
|
||||
|
||||
def _prev_date(table: str, before: str):
|
||||
df = db.read_mysql(
|
||||
"factor", f"SELECT MAX(trade_date) d FROM {table} "
|
||||
f"WHERE trade_date < %s", (before,))
|
||||
v = None if df.empty else df.iloc[0, 0]
|
||||
return None if v is None or pd.isna(v) else pd.Timestamp(v).date().isoformat()
|
||||
|
||||
|
||||
def _names() -> dict:
|
||||
out = {}
|
||||
pools = db.read_pg("SELECT members FROM industry_pools")
|
||||
for _, r in pools.iterrows():
|
||||
ms = r["members"]
|
||||
if isinstance(ms, str):
|
||||
ms = json.loads(ms)
|
||||
for m in ms or []:
|
||||
ts, name = (m or {}).get("ts_code"), (m or {}).get("name")
|
||||
if ts and name:
|
||||
out.setdefault(common.to_prefix(ts), str(name))
|
||||
return out
|
||||
|
||||
|
||||
def _evidence(ds: str):
|
||||
"""每股最强一条传导证据：主题、源数、已动比例；另返回涉及的行情快照日。"""
|
||||
tr = db.read_pg(
|
||||
"SELECT ts_code, target, n_sources, moved_ratio, mkt_trade_date "
|
||||
"FROM v_factor_transmission WHERE scan_date = %s", (ds,))
|
||||
if tr.empty:
|
||||
return {}, set()
|
||||
tr["k"] = tr["ts_code"].map(common.to_prefix)
|
||||
tr["n_sources"] = pd.to_numeric(tr["n_sources"], errors="coerce").fillna(0)
|
||||
tr["moved_ratio"] = pd.to_numeric(tr["moved_ratio"], errors="coerce").fillna(0)
|
||||
tr["strength"] = tr["n_sources"] * (1.0 - tr["moved_ratio"])
|
||||
tr = tr.sort_values("strength", ascending=False).drop_duplicates("k")
|
||||
ev = {r.k: (str(r.target), int(r.n_sources), float(r.moved_ratio))
|
||||
for r in tr.itertuples()}
|
||||
days = {str(x) for x in tr["mkt_trade_date"].dropna().unique()}
|
||||
return ev, days
|
||||
|
||||
|
||||
def _tier_label(score: float) -> str:
|
||||
return {0: "无传导", 1: "弱传导", 2: "强传导"}.get(
|
||||
int((score - 190.0) // 20), "?")
|
||||
|
||||
|
||||
def _pct(v) -> str:
|
||||
return "—" if v is None or pd.isna(v) else f"{v:+.0%}"
|
||||
|
||||
|
||||
def _num(v) -> str:
|
||||
return "—" if v is None or pd.isna(v) else f"{v:.2f}"
|
||||
|
||||
|
||||
def generate(date: str | None = None, top: int = 20, obs_top: int = 10) -> str:
|
||||
ds = date or _latest_date("t_factor_akg_score")
|
||||
if not ds:
|
||||
raise SystemExit("t_factor_akg_score 还没有数据——先 build akg_score。")
|
||||
score = _factor("t_factor_akg_score", ds)
|
||||
gate = _factor("t_factor_akg_gate", ds)
|
||||
if score.empty or gate.empty:
|
||||
raise SystemExit(f"{ds} 缺 akg_score / akg_gate——先 build 该日再出计划。")
|
||||
upside = _factor("t_factor_akg_upside", ds)
|
||||
hd = _latest_date("t_factor_akg_heat", ds)
|
||||
heat = _factor("t_factor_akg_heat", hd) if hd else pd.Series(dtype=float)
|
||||
names = _names()
|
||||
ev, mkt_days = _evidence(ds)
|
||||
|
||||
main = score[score >= _MAIN_MIN].sort_values(ascending=False)
|
||||
obs = score[score < _MAIN_MIN].sort_values(ascending=False)
|
||||
|
||||
L = [f"# 每日选股计划 · {ds}", ""]
|
||||
L.append(f"主榜 {len(main)} 只 / 观察档 {len(obs)} 只 / 全池档位覆盖 {len(gate)} 只。")
|
||||
stale = sorted(d for d in mkt_days if d != ds)
|
||||
if stale:
|
||||
L.append(f"⚠️ 本日传导用的行情快照 = {'、'.join(stale)}（T−1 口径："
|
||||
f"\"谁已经动了\"看的是上个交易日收盘；拍点方案定版前均如此）。")
|
||||
L.append("")
|
||||
|
||||
L.append(f"## 主榜 Top {min(top, len(main))}（有券商预期、目标价不低于现价）")
|
||||
L.append("")
|
||||
L.append("| # | 代码 | 名称 | 总分 | 档位 | 传导证据 | 热度 | 预期空间 |")
|
||||
L.append("|---|------|------|------|------|----------|------|----------|")
|
||||
for i, (k, s) in enumerate(main.head(top).items(), 1):
|
||||
e = ev.get(k)
|
||||
etxt = f"{e[0]}（{e[1]} 源，已动 {e[2]:.0%}）" if e else "—"
|
||||
L.append(f"| {i} | {k} | {names.get(k, '—')} | {s:.1f} | {_tier_label(s)} "
|
||||
f"| {etxt} | {_num(heat.get(k))} | {_pct(upside.get(k))} |")
|
||||
L.append("")
|
||||
|
||||
L.append(f"## 观察档 Top {min(obs_top, len(obs))}"
|
||||
f"（无券商预期、但在传导链上——没有估值锚，置信度低）")
|
||||
L.append("")
|
||||
L.append("| # | 代码 | 名称 | 分 | 传导证据 | 热度 |")
|
||||
L.append("|---|------|------|----|----------|------|")
|
||||
for i, (k, s) in enumerate(obs.head(obs_top).items(), 1):
|
||||
e = ev.get(k)
|
||||
etxt = f"{e[0]}（{e[1]} 源，已动 {e[2]:.0%}）" if e else "—"
|
||||
L.append(f"| {i} | {k} | {names.get(k, '—')} | {s:.1f} "
|
||||
f"| {etxt} | {_num(heat.get(k))} |")
|
||||
L.append("")
|
||||
|
||||
L.append("## 今日升降档")
|
||||
L.append("")
|
||||
prev_ds = _prev_date("t_factor_akg_gate", ds)
|
||||
if not prev_ds:
|
||||
L.append("（没有更早的档位表可比，升降档从下一个交易日开始。）")
|
||||
else:
|
||||
prev = _factor("t_factor_akg_gate", prev_ds)
|
||||
both = pd.concat([prev.rename("prev"), gate.rename("cur")], axis=1)
|
||||
both = both.fillna(-1.0) # -1 = 当日不在面板
|
||||
up = both[both["cur"] > both["prev"]]
|
||||
down = both[both["cur"] < both["prev"]]
|
||||
lab = {-1.0: "池外", 0.0: "不采纳", 1.0: "观察档", 2.0: "主榜"}
|
||||
L.append(f"对比 {prev_ds}：升档 {len(up)} 只，降档 {len(down)} 只。"
|
||||
f"升档=拿到新锚（首次覆盖 / 新进传导链），本身就是值得看的信号。")
|
||||
|
||||
def _rows(d: pd.DataFrame, cap: int = 15):
|
||||
lines = []
|
||||
for k, r in d.iterrows():
|
||||
lines.append(f"- {k} {names.get(k, '')}："
|
||||
f"{lab.get(r['prev'], '?')} → {lab.get(r['cur'], '?')}")
|
||||
if len(lines) >= cap:
|
||||
lines.append(f"- ……共 {len(d)} 只，其余见档位表")
|
||||
break
|
||||
return lines
|
||||
|
||||
if not up.empty:
|
||||
L.append("")
|
||||
L.append("**升档**：")
|
||||
L += _rows(up.sort_values("cur", ascending=False))
|
||||
if not down.empty:
|
||||
L.append("")
|
||||
L.append("**降档**：")
|
||||
L += _rows(down.sort_values("prev", ascending=False))
|
||||
L.append("")
|
||||
L.append("---")
|
||||
L.append("口径：主榜分 = 200 + 传导档位×20 + 组内分（还没热、还便宜）；"
|
||||
"观察档分 = 100 + 0.6z(传导) + 0.4z(−热度)。研究用途，不构成投资建议。")
|
||||
|
||||
text = "\n".join(L)
|
||||
os.makedirs("data/plan", exist_ok=True)
|
||||
out = f"data/plan/plan_{ds}.md"
|
||||
with open(out, "w", encoding="utf-8") as f:
|
||||
f.write(text + "\n")
|
||||
print(text)
|
||||
print(f"\n已写入 {out}")
|
||||
return out
|
||||
14
run.py
14
run.py
|
|
@ -5,6 +5,7 @@
|
|||
python run.py probe # G1 体检（只读，详见 probe.py）
|
||||
python run.py freeze [--date D] # 输入冻结（G0.5，详见 freeze.py）
|
||||
python run.py tracks # 赛道覆盖体检 + 成员表快照（G2）
|
||||
python run.py plan [--date D] [--top N] # 每日选股计划（R4，读已落库因子表）
|
||||
python run.py register # 注册全部因子到 factor_metadata
|
||||
python run.py build all --mode history --start 2024-01-01 --end 2025-12-31
|
||||
python run.py build akg_heat --mode daily --date 2026-07-24
|
||||
|
|
@ -94,9 +95,9 @@ _META = {
|
|||
"转正后再交赛道成员), 1=观察档(无券商覆盖但在图谱传导链上, 无估值锚, "
|
||||
"低置信), 0=不采纳(两锚皆无, 或upside<0)。全池出行, 让平台看得见门槛"),
|
||||
"akg_score": ("astock-kg 景气度漏斗",
|
||||
"先档后分: 主榜=200+传导档位x10+组内分(0.6z(-热度)+0.4z(upside), "
|
||||
"两段式07-30拍板); 观察档=100+0.6z(传导)+0.4z(-热度)。"
|
||||
"仅gate>0出行, 数值直接可排序"),
|
||||
"先档后分: 主榜=200+传导档位x20+组内分(0.6z(-热度)+0.4z(upside), "
|
||||
"两段式07-30拍板, 组内分clip±9.9故档间不重叠); "
|
||||
"观察档=100+0.6z(传导)+0.4z(-热度)。仅gate>0出行, 数值直接可排序"),
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -140,6 +141,10 @@ def main():
|
|||
sub.add_parser("views")
|
||||
sub.add_parser("register")
|
||||
sub.add_parser("tracks") # 赛道覆盖体检 + confirmed 成员表快照（G2）
|
||||
pl = sub.add_parser("plan") # 每日选股计划（R4）
|
||||
pl.add_argument("--date", help="默认取 score 表最新日")
|
||||
pl.add_argument("--top", type=int, default=20, help="主榜条数")
|
||||
pl.add_argument("--obs-top", type=int, default=10, help="观察档条数")
|
||||
p = sub.add_parser("probe")
|
||||
p.add_argument("--section", choices=["all", "pools", "price", "upside", "corr"],
|
||||
default="all",
|
||||
|
|
@ -171,6 +176,9 @@ def main():
|
|||
freeze.snapshot(a.date)
|
||||
elif a.cmd == "register":
|
||||
cmd_register()
|
||||
elif a.cmd == "plan":
|
||||
import plan
|
||||
plan.generate(a.date, a.top, a.obs_top)
|
||||
elif a.cmd == "tracks":
|
||||
import tracks
|
||||
tracks.coverage_report()
|
||||
|
|
|
|||
16
tracks.py
16
tracks.py
|
|
@ -37,7 +37,8 @@ def load_yml(path: str | None = None) -> dict:
|
|||
return yaml.safe_load(f)
|
||||
|
||||
|
||||
def resolve_members(only_confirmed: bool = True, path: str | None = None):
|
||||
def resolve_members(only_confirmed: bool = True, path: str | None = None,
|
||||
dedup: bool = True):
|
||||
"""yml + industry_pools 当前态 → (成员表, 未命中主题清单)。
|
||||
|
||||
成员表列：ts_code, name, track, status, theme。同股同赛道多主题只留一行，
|
||||
|
|
@ -69,9 +70,10 @@ def resolve_members(only_confirmed: bool = True, path: str | None = None):
|
|||
if ts:
|
||||
rows.append((ts, m.get("name"), tr["name"],
|
||||
tr.get("status"), t))
|
||||
df = (pd.DataFrame(rows, columns=["ts_code", "name", "track",
|
||||
"status", "theme"])
|
||||
.drop_duplicates(["ts_code", "track"]))
|
||||
df = pd.DataFrame(rows, columns=["ts_code", "name", "track",
|
||||
"status", "theme"])
|
||||
if dedup:
|
||||
df = df.drop_duplicates(["ts_code", "track"])
|
||||
return df, missing
|
||||
|
||||
|
||||
|
|
@ -88,9 +90,11 @@ def snapshot(only_confirmed: bool = True, path: str | None = None):
|
|||
|
||||
def coverage_report(path: str | None = None) -> pd.DataFrame:
|
||||
"""赛道覆盖体检（G1 清单的首件事）：逐赛道成员数与主题命中率，含 candidate。"""
|
||||
df, missing = resolve_members(only_confirmed=False, path=path)
|
||||
# dedup=False：主题命中率要在去重前数——成员完全被同赛道更早主题收进来的
|
||||
# 主题（如卫星导航之于卫星），去重后一行不剩，会被误计成"没命中"（07-30 实测）。
|
||||
df, missing = resolve_members(only_confirmed=False, path=path, dedup=False)
|
||||
d = load_yml(path)
|
||||
print("赛道覆盖体检（industry_pools 当前态；成员数已去重）：")
|
||||
print("赛道覆盖体检（industry_pools 当前态；成员数已去重，主题命中按去重前算）：")
|
||||
for tr in d.get("tracks") or []:
|
||||
sub = df[df["track"] == tr["name"]]
|
||||
n_theme = len(tr.get("kg_themes") or [])
|
||||
|
|
|
|||
Loading…
Reference in New Issue