# Ex-ante daily regime labels L1..L8 (PREREG). Everything uses data known by 09:15 of the trade day.
import sys; sys.path.insert(0,'/workspace/research/popular-strategies/scripts'); from plib import *
import os
FR='/workspace/nifty-market-games-knowledge/research/factor-regression/data'
S,PC=load_spot(); days=sorted(S); D=daily_table(S,PC)
# daily OHLC: official (Upstox daily) 2022+; yahoo 2021 only for warm-up
off=pd.read_parquet('/workspace/research/common/data/nifty_daily_hist.parquet')
u=pd.read_csv('/workspace/research/flow-forensics-2026-10/data/1m/NIFTY_day.csv'); u['date']=pd.to_datetime(u.ts.str[:10])
off=pd.concat([off,u[['date','open','high','low','close']]]).drop_duplicates('date',keep='last')
y=pd.read_parquet(f'{FR}/raw_NIFTY.parquet'); y.columns=[c.lower() for c in y.columns]; y=y.reset_index().rename(columns={'Date':'date'}); y['date']=pd.to_datetime(y.date)
y=y[(y.date>='2020-06-01')&(y.date<off.date.min())][['date','open','high','low','close']]
DD=pd.concat([y,off]).sort_values('date').set_index('date'); DD=DD[DD.index<=LAST]
c=DD.close; rng_=DD.high-DD.low
lab=pd.DataFrame(index=DD.index)
pcl=c.shift(1); lab['prev_close']=pcl
lab['dma20']=c.rolling(20).mean().shift(1); lab['dma50']=c.rolling(50).mean().shift(1)
lab['L2_DMA_trend']=np.where((pcl>lab.dma20)&(pcl>lab.dma50),'up',np.where((pcl<lab.dma20)&(pcl<lab.dma50),'down','mixed')); lab.loc[lab.dma50.isna(),'L2_DMA_trend']=np.nan
er=(c-c.shift(20)).abs()/c.diff().abs().rolling(20).sum(); er=er.shift(1); lab['er20']=er
lab['L3_ER20']=pd.cut(er,[-1,0.2,0.4,2],labels=['range<0.2','mid0.2-0.4','trend>0.4']).astype(str).replace('nan',np.nan)
pr=rng_.shift(1); avg=rng_.shift(2).rolling(20).mean(); lab['pdr_ratio']=pr/avg
lab['L4_PDR']=pd.cut(lab.pdr_ratio,[0,0.8,1.2,99],labels=['compressed<0.8','normal','expanded>1.2']).astype(str).replace('nan',np.nan)
lab['L5_NR7']=np.where(pr<=rng_.shift(1).rolling(7).min(),'NR7','no'); lab.loc[pr.isna(),'L5_NR7']=np.nan
g=(DD.open-pcl)/pcl; lab['gap_pct']=g*100
lab['L6_Gap']=pd.cut(g.abs()*100,[-1,0.25,0.75,99],labels=['flat<0.25%','medium','large>0.75%']).astype(str).replace('nan',np.nan)
# VIX (daily open == prior close for India VIX)
v=pd.read_parquet(f'{FR}/raw_INDIAVIX.parquet'); v=v.reset_index(); v.columns=['date']+[x.lower() for x in v.columns[1:]]; v['date']=pd.to_datetime(v.date)
vu=pd.read_csv('/workspace/research/flow-forensics-2026-10/data/1m/INDIAVIX_day.csv'); vu['date']=pd.to_datetime(vu.ts.str[:10])
V=pd.concat([v[['date','close']],vu[['date','close']]]).drop_duplicates('date',keep='last').set_index('date').close.sort_index()
lab['vix_open']=V.reindex(lab.index).shift(1).values  # prior VIX close = VIX at 09:15 open
lab['L1_VIX']=pd.cut(lab.vix_open,[0,13,16,999],labels=['<13','13-16','>16']).astype(str).replace('nan',np.nan)
# expiry
opt(); import lib; EX=set(lib._EXPS); td=[d for d in days]
def thursday_exp(d):
    # weekly expiry: Thursday (Tuesday from 2025-09-01), shifted to previous trading day if holiday
    tgt=3 if d<pd.Timestamp('2025-09-01') else 1
    wk=d-pd.Timedelta(days=d.weekday()); t=wk+pd.Timedelta(days=tgt)
    while t not in set(td) and t>=wk: t-=pd.Timedelta(days=1)
    return t
exps=set()
for d in td:
    if d>=pd.Timestamp('2024-10-21'): continue
    exps.add(thursday_exp(d))
exps|={e for e in EX if e<=LAST}
tds=sorted(td); pos={d:i for i,d in enumerate(tds)}
def ex_lab(d):
    if d in exps: return 'expiry_day'
    i=pos[d]
    if i+1<len(tds) and tds[i+1] in exps: return 'day_before'
    return 'other'
lab=lab.reindex(tds); lab['L7_Expiry']=[ex_lab(d) for d in tds]
f='/workspace/research/popular-strategies/data/fii_index_fut_long_pct.csv'
if os.path.exists(f):
    F=pd.read_csv(f,parse_dates=['date']).set_index('date').fii_long_pct.sort_index()
    # prior trading day's participant file (published after close) -> known before next open
    prev=[]; 
    for d in tds:
        s=F[F.index<d]; prev.append(s.iloc[-1] if len(s) and (d-s.index[-1]).days<=5 else np.nan)
    lab['fii_long_pct']=prev
    lab['L8_FII_long']=pd.cut(lab.fii_long_pct,[0,25,45,100],labels=['<25%','25-45%','>45%']).astype(str).replace('nan',np.nan)
lab.index.name='date'; lab.to_csv('/workspace/research/popular-strategies/results/regime_labels.csv')
LB=[c for c in lab.columns if c[:2] in ('L1','L2','L3','L4','L5','L6','L7','L8')]
for c_ in LB: print(c_, lab[c_].value_counts(dropna=False).to_dict())