"""Independent implementation from Cadosch et al. eqs (1)-(6), Table 2. Does not import the worker's code, outputs, or prior referee. """ import os for key in ('OPENBLAS_NUM_THREADS','OMP_NUM_THREADS','MKL_NUM_THREADS'): os.environ[key]='1' import json, itertools, subprocess, time import numpy as np from scipy.integrate import odeint from scipy.optimize import minimize from scipy.special import expit, gammaln import statsmodels.api as sm started=time.monotonic() years=np.arange(2007,2011)+.5 n=np.array([11,226,378,454]); k=np.array([3,81,135,191]) external_years=np.arange(2021,2025)+.5 beta=.8/.98; gamma=.8; mu=.12 initial=[.98,.02,0.,0.] def derivatives(x,t,tau): s,w,a,r=x allinfected=w+a+r return [-beta*s*allinfected+gamma*allinfected+(1-mu)*tau*w, beta*s*w-gamma*w-tau*w, mu*tau*w-gamma*a, beta*s*(a+r)-gamma*r] def resistance(t,tau,introduction,precise=False): t=np.asarray(t) opts={'rtol':1e-10,'atol':1e-13} if precise else {'rtol':1e-6,'atol':1e-6} x=odeint(derivatives,initial,np.r_[introduction,t],args=(tau,),**opts)[1:] return (x[:,2]+x[:,3])/x[:,1:].sum(axis=1) def transform(v): return np.exp(v[0]),1990+expit(v[1])*(years[0]-1990) def objective(v): tau,introduction=transform(v) p=resistance(years,tau,introduction) if np.any(p<=0) or np.any(p>=1):return float('inf') return -float(np.sum(gammaln(n+1)-gammaln(k+1)-gammaln(n-k+1)+k*np.log(p)+(n-k)*np.log1p(-p))) # First run: the paper's standard Nelder-Mead procedure and transformed start. fit=minimize(objective,[np.log(.4),0.],method='Nelder-Mead') assert fit.success,fit own_tau,own_intro=transform(fit.x) # Also evaluate frozen source parameter values as literal data, rather than substituting our refit. frozen_tau=.129069567169382; frozen_intro=1995.5803585762 source=resistance(external_years,frozen_tau,frozen_intro,precise=True) refitted=resistance(external_years,own_tau,own_intro,precise=True) # Independent analytic identity derived by dividing (3)+(4) by (2): # r'=tau*(r+mu), r(T)=0; p=r/(1+r). This is a cross-check, not the fitter. r=mu*np.expm1(frozen_tau*(external_years-frozen_intro)); analytic=r/(1+r) assert np.max(abs(source-analytic))<1e-8 assert np.max(abs(source-refitted))<.002 # Standard grouped-binomial GLM; only the four printed historical training rows. x=sm.add_constant(years-2010) glm=sm.GLM(np.column_stack([k,n-k]),x,family=sm.families.Binomial()).fit() simple=glm.predict(sm.add_constant(external_years-2010)) trivial=np.repeat(k[-1]/n[-1],4) # Exact SSI numerator/denominator, and printed MiBa centers, transcribed from primary sources. ssi=np.array([274/438,311/436]); miba=np.array([.606,.632,.636,.65]) # Integer denominator audit for the 2021 overall and sex-specific counts/rounding. possible=lambda positive,pct:[den for den in range(positive,2000) if pct-.05<=100*positive/den