Function cell

scripts/lambda_floor_ablation.py:50–77

(lam, floor)

logN = np.array([math.log10(CFG[m["model"]]["params_B"]) for m in opens])

def cell(lam, floor):
    a = np.array([acc(m["tier_stats"], lam, floor) for m in opens])
    # forward fit A = alpha*log10N + beta
    alpha, beta, r, _, _ = stats.linregress(logN, a)
    resid = a - (alpha*logN + beta)
    se = math.sqrt(float(np.sum(resid**2))/max(len(a)-2,1))
    pi = 10**(1.645*se/abs(alpha)) if alpha else float("inf")
    # estimator slope S in log10N = S*acc + I (what ikp_estimate reports; critique's 6.79)
    S, I, rS, _, _ = stats.linregress(a, logN)
    # LOO
    folds=[]; n=len(opens)
    for i in range(n):
        mk=np.ones(n,bool); mk[i]=False
        al,be,_,_,_=stats.linregress(logN[mk],a[mk])
        if al>0: folds.append(10**abs((a[i]-be)/al - logN[i]))
    med=float(np.median(folds)); w2=float(np.mean(np.array(folds)<=2)); w3=float(np.mean(np.array(folds)<=3))
    # estimates
    est={}
    for name in SPOT:
        m=by.get(name)
        if m and alpha>0:
            av=acc(m["tier_stats"], lam, floor)
            e=10**((av-beta)/alpha)
            est[name]={"acc":av,"est_B":e,"lo":e/pi,"hi":e*pi}
        else: est[name]=None
    return {"lambda":lam,"floor":floor,"n":n,"R2":r**2,"alpha_pp":alpha*100,
            "estimator_slope_S":S,"pi_factor":pi,"loo_med":med,"within2":w2,"within3":w3,
            "estimates":est}

results=[cell(l,f) for l,f in product(LAMBDAS,[True,False])]
OUTJSON.parent.mkdir(parents=True, exist_ok=True)