# Premium-support ("black box") variants V1-V4 on the ATM option premium chart; exits X_OLD and X_T50 (PREREG PREMIUM-SUPPORT section).
import sys,time; sys.path.insert(0,'/workspace/research/popular-strategies/scripts'); from plib import *
t0=time.time(); S,PC=load_spot(); OPT=opt(); RS=f'{L}/results'
DAYS=[d for d in sorted(S) if pd.Timestamp('2024-10-21')<=d<=LAST and nexp(d) is not None]
ALLD=sorted(S); PREV={d:ALLD[i-1] for i,d in enumerate(ALLD) if i>0}
T_END=hm2i('15:15'); W0=hm2i('09:44'); W1=hm2i('14:29'); W1B=hm2i('13:29'); SL=0.5
def prem5(oa):
    o,h,l,c=oa; P=np.full((4,72),np.nan)
    for k in range(72):
        s=slice(5*k,5*k+5); cc=c[s]
        if np.all(np.isnan(cc)): continue
        idx=np.where(~np.isnan(cc))[0]; P[0,k]=o[s][idx[0]]; P[1,k]=np.nanmax(h[s]); P[2,k]=np.nanmin(l[s]); P[3,k]=cc[idx[-1]]
    return P
_p5={}
def chart(d,e,typ,K):
    key=(d,e,typ,K)
    if key not in _p5:
        oa=oarr(d,e,typ,K); _p5[key]=(oa,prem5(oa)) if oa is not None else (None,None)
    return _p5[key]
def spot5c(a): 
    c=a[3]; return np.array([c[5*k+4] if not np.isnan(c[5*k+4]) else (c[5*k:5*k+5][~np.isnan(c[5*k:5*k+5])][-1] if (~np.isnan(c[5*k:5*k+5])).any() else np.nan) for k in range(72)])
def tol(x): return max(2.0,0.03*x)
def pdl(d,e,typ,K):
    p=PREV.get(d); 
    if p is None: return np.nan
    O=OPT.get((p,e))
    if O is None or ('low',typ,K) not in O.columns: return np.nan
    v=O[('low',typ,K)].values; return np.nanmin(v) if np.isfinite(v).any() else np.nan
def touch_ok(P,k,lvl):
    if not np.isfinite(lvl) or k<1: return False
    t=tol(lvl); return (lvl-t<=P[2,k]<=lvl+t) and (P[3,k-1]>lvl+t)
def scan(d,var):
    e=nexp(d); a=S[d]; sc=spot5c(a); hits=[]
    for j in range(1,72):
        m=5*j+4
        if m<W0 or m>(W1B if var=='V4' else W1): continue
        for typ in ('CE','PE'):
            if var=='V4':
                if np.isnan(sc[j]): continue
                K=float(round(sc[j]/50)*50); oa,P=chart(d,e,typ,K)
                if P is None or np.isnan(P[3,j]): continue
                ph=np.nanmax(P[1,:j]) if np.isfinite(P[1,:j]).any() else np.nan
                if np.isfinite(ph) and P[3,j]>ph: hits.append((m,typ,K,P[2,j]-1,ph))
                continue
            k=j-1
            if np.isnan(sc[k]): continue
            K=float(round(sc[k]/50)*50); oa,P=chart(d,e,typ,K)
            if P is None or np.isnan(P[3,j]) or np.isnan(P[2,k]): continue
            if var=='V1': lvl=pdl(d,e,typ,K)
            elif var=='V2':
                lvl=np.nan
                for p in range(k-3,2,-1):
                    if p+2>=k or p-2<0: continue
                    w=P[2,p-2:p+3]
                    if np.isfinite(w).all() and P[2,p]==w.min() and (w<P[2,p]).sum()==0 and (w==P[2,p]).sum()==1: lvl=P[2,p]; break
            else:
                lw=P[2,k]; lvl=round(lw/25)*25 if 40<=lw<=400 else np.nan
                if lvl==0: lvl=np.nan
            if touch_ok(P,k,lvl) and P[3,j]>P[1,k] and P[3,j]>lvl: hits.append((m,typ,K,P[2,k]-1,lvl))
        if hits: break
    if not hits: return None
    if len({h[1] for h in hits})>1: return 'both'
    return dict(zip(['m','typ','K','stop','level'],hits[0]))+dict(e=e) if False else dict(m=hits[0][0],typ=hits[0][1],K=hits[0][2],stop=hits[0][3],level=hits[0][4],e=e)
def run_exit(d,oa,m,ex,Rp=None):
    raw=oa[0,m+1] if m+1<NM else np.nan
    if np.isnan(raw): return None
    fill=raw+SL; sells=[]; why='time'; xx=None
    if ex=='X_OLD':
        stop=fill-20; t1=t2=False; peak=fill; rem=1.0
        for x in range(m+1,T_END):
            op,hi,lo,cl=oa[:,x]
            if np.isnan(cl): continue
            if lo<=stop: sells.append((rem,min(op,stop)-SL if x>m+1 else stop-SL)); rem=0; why='stop' if not t1 else 'lock'; xx=x; break
            if not t1 and hi>=fill+20: sells.append((0.4,fill+20-SL)); rem-=0.4; t1=True; stop=max(stop,fill+10)
            if not t2 and hi>=fill+40: sells.append((0.3,fill+40-SL)); rem-=0.3; t2=True; stop=max(stop,fill+25)
            if hi>=fill+60: sells.append((rem,fill+60-SL)); rem=0; why='target'; xx=x; break
            peak=max(peak,hi)
            if t2: stop=max(stop,peak-15)
    else:
        Rp=min(max(Rp,5.0),30.0); stop=raw-Rp; peak=0.0; rem=1.0
        for x in range(m+1,T_END):
            op,hi,lo,cl=oa[:,x]
            if np.isnan(cl): continue
            if lo<=stop: sells.append((1.0,(min(op,stop) if x>m+1 else stop)-SL)); rem=0; why='stop'; xx=x; break
            pr=cl-raw; peak=max(peak,pr)
            if peak>=2*Rp and pr<=0.5*peak:
                px,y=next_px(oa,x+1)
                if y is not None: sells.append((1.0,px-SL)); rem=0; why='trail'; xx=y; break
    if rem>1e-9:
        px,y=next_px(oa,T_END)
        if y is None:
            v=np.where(~np.isnan(oa[3,:T_END+10]))[0]; v=v[v>m]
            if not len(v): return None
            px=oa[3,v[-1]]; y=v[-1]
        sells.append((rem,px-SL)); xx=y
    g=sum(f*p for f,p in sells)-fill; c_=charges(d,fill,sells,True)
    return dict(raw=raw,fill=fill,gross=g,cost=c_,pnl=g-c_,why=why,exit_min=i2hm(xx),n_orders=1+len(sells))
LAB=pd.read_csv(f'{RS}/regime_labels.csv',parse_dates=['date']).set_index('date')
rows=[]; skipped=dict(both=0,nobar=0)
for var in ('V1','V2','V3','V4'):
    for d in DAYS:
        s=scan(d,var)
        if s is None: continue
        if s=='both': skipped['both']+=1; continue
        oa,_=chart(d,s['e'],s['typ'],s['K']); raw=oa[0,s['m']+1] if s['m']+1<NM else np.nan
        if np.isnan(raw): skipped['nobar']+=1; continue
        Rp=raw-s['stop']
        for ex in ('X_OLD','X_T50'):
            z=run_exit(d,oa,s['m'],ex,Rp)
            if z: rows.append(dict(var=var,ex=ex,date=d,hm=i2hm(s['m']),typ=s['typ'],K=s['K'],expiry=s['e'],level=s['level'],Rp=min(max(Rp,5),30),split=split(d),**z))
    print(var,'done',round(time.time()-t0),flush=True)
T=pd.DataFrame(rows); T.to_csv(f'{RS}/prem_trades.csv',index=False); print(T.groupby(['var','ex','split']).pnl.agg(['size','mean']).round(2),skipped,flush=True)
# random benchmark
rng=np.random.default_rng(2124); pool={sp:[d for d in DAYS if split(d)==sp] for sp in ('train','test')}; rr=[]
for r in T.itertuples():
    for k in range(30):
        d=pool[r.split][rng.integers(len(pool[r.split]))]; a=S[d]; m=hm2i(r.hm); sc=a[3,m]
        if np.isnan(sc): continue
        e=nexp(d); oa=oarr(d,e,r.typ,float(round(sc/50)*50))
        if oa is None: continue
        z=run_exit(d,oa,m,r.ex,r.Rp)
        if z: rr.append(dict(var=r.var,ex=r.ex,src_date=r.date,rep=k,date=d,split=r.split,pnl=z['pnl']))
pd.DataFrame(rr).to_csv(f'{RS}/prem_random.csv',index=False); print('random',len(rr),round(time.time()-t0))