import kit,numpy as np,sys,datetime as dt
sys.stdout.reconfigure(encoding='utf-8',errors='replace')
from sklearn.ensemble import HistGradientBoostingClassifier
sym,tf,sl,tp,H,mn=("EURUSD","16385",2,3,48,1999)
t,o,h,l,c,v,sp=kit.load_rates(sym,tf)
k=np.array([dt.datetime.fromtimestamp(x,dt.UTC).year for x in t])>=mn
t,o,h,l,c,v,sp=[a[k] for a in (t,o,h,l,c,v,sp)]
X,names,a=kit.features(t,o,h,l,c,v)
diffs=np.abs(np.diff(np.unique(np.round(c,6)))); tick=np.median(diffs[diffs>0])
y,valid=kit.barrier_vec(h,l,c,a,sl,tp,H,np.median(sp)*tick)
n=len(y); idx=np.arange(n); be=100*sl/(sl+tp)
prob=np.full((n,3),np.nan); fold=np.full(n,-1)
bounds=np.linspace(int(n*0.35),n,6).astype(int)
for f in range(5):
    te0,te1=bounds[f],bounds[f+1]
    te=(idx>=te0)&(idx<te1)&valid; tr=(idx<te0-H)&valid
    m=HistGradientBoostingClassifier(max_iter=200,learning_rate=0.06,max_depth=6,
        l2_regularization=1.0,random_state=0).fit(X[tr],y[tr])
    prob[te]=m.predict_proba(X[te]); fold[te]=f
print(f"EURUSD H1 {sl}:{tp} h{H}  break-even {be:.1f}%   bars={n}  ({dt.datetime.fromtimestamp(t[0],dt.UTC).year}-{dt.datetime.fromtimestamp(t[-1],dt.UTC).year})")
print("ALL thresholds, no selection:\n")
print(f"  {'thr':>5}{'trades':>8}{'win%':>8}{'edge':>8}{'sigma':>8}   per-fold win%")
for thr in (0.35,0.40,0.45,0.50,0.55,0.60):
    tr_i=[];j=0
    while j<n:
        if fold[j]<0 or not np.isfinite(prob[j,0]): j+=1; continue
        p=prob[j];pr=int(np.argmax(p))
        if pr==2 or float(np.max(p))<thr: j+=1; continue
        tr_i.append((pr==y[j],fold[j]));j+=H
    if len(tr_i)<50: print(f"  {thr:>5.2f}{len(tr_i):>8}   too few"); continue
    w=np.array([x[0] for x in tr_i]);fo=np.array([x[1] for x in tr_i])
    wr=100*w.mean();nT=len(w);se=100*np.sqrt((be/100)*(1-be/100)/nT)
    pf=" ".join(f"{100*w[fo==f].mean():5.1f}" for f in range(5) if (fo==f).sum()>=15)
    print(f"  {thr:>5.2f}{nT:>8}{wr:>8.2f}{wr-be:>+8.2f}{(wr-be)/se:>+8.2f}   {pf}")
print("\nsigma is vs break-even on INDEPENDENT trades. 6 thresholds tested here, 48 across the")
print("whole sweep - so a single +1.5s reading is what selection alone produces.")