"""PeerJ 8913 eqs 1--6 and Table 2; no claimant/referee code imports.""" import csv,itertools,json,os,subprocess from pathlib import Path import numpy as np from scipy.integrate import solve_ivp from scipy.optimize import minimize from scipy.special import expit source=Path('friends_note_victor_1104') t=np.array([2007.5,2008.5,2009.5,2010.5]) n=np.array([11,226,378,454]); k=np.array([3,81,135,191]) dates=np.array([2021.5,2022.5,2023.5,2024.5]) def predict(tau,intro,dates): def rhs(t,y): s,w,a,r=y; beta=.8/.98; gamma=.8; mu=.12 infected=w+a+r return [-beta*s*infected+gamma*infected+(1-mu)*tau*w, beta*s*w-(gamma+tau)*w,mu*tau*w-gamma*a, beta*s*(a+r)-gamma*r] sol=solve_ivp(rhs,(intro,max(dates)),[.98,.02,0,0],t_eval=dates, method='LSODA',rtol=1e-10,atol=1e-13) if not sol.success: raise RuntimeError(sol.message) return (sol.y[2]+sol.y[3])/sol.y[1:].sum(axis=0) def loss(raw): p=predict(np.exp(raw[0]),1990+17.5*expit(raw[1]),t) return -np.sum(k*np.log(p)+(n-k)*np.log1p(-p)) fit=minimize(loss,[np.log(.4),0],method='Nelder-Mead', options={'maxiter':500,'xatol':1e-9,'fatol':1e-10}) tau_fit=np.exp(fit.x[0]); intro_fit=1990+17.5*expit(fit.x[1]) with (source/'results/standard/parameters.csv').open() as f: frozen=next(row for row in csv.DictReader(f) if row['country']=='Denmark') tau=float(frozen['tau']); intro=float(frozen['intro']) method=predict(tau,intro,dates) # Derive r=(I_A+I_T)/I_S from equations: r'=tau*(r+mu), r(T)=0. ratio=.12*np.expm1(tau*(dates-intro)); scalar=ratio/(1+ratio) # Ordinary binomial logistic Newton/IRLS, four historical rows only. X=np.column_stack([np.ones(4),t-2009.5]); coef=np.zeros(2) for _ in range(30): p=expit(X@coef) step=np.linalg.solve(X.T@((n*p*(1-p))[:,None]*X),X.T@(k-n*p)) coef+=step if max(abs(step))<1e-12: break simple=expit(np.column_stack([np.ones(4),dates-2009.5])@coef) trivial=np.full(4,k[-1]/n[-1]) ssi=np.array([274/438,311/436]); miba=np.array([.606,.632,.636,.65]) out={'training':{'time':t.tolist(),'n':n.tolist(),'k':k.tolist()}, 'frozen_parameters':{'tau':tau,'intro':intro}, 'independent_fit':{'tau':float(tau_fit),'intro':float(intro_fit),'success':bool(fit.success)}, 'max_ode_scalar_difference':float(max(abs(method-scalar))), 'logistic_coefficients':coef.tolist(), 'predictions':{'date':dates.tolist(),'method':method.tolist(), 'simple':simple.tolist(),'trivial':trivial.tolist(), 'refit':predict(tau_fit,intro_fit,dates).tolist()},'scores':{}} baselines={} for name,obs in [('SSI',ssi),('MiBa',miba)]: errors=[float(np.mean(abs(pred[:len(obs)]-obs))) for pred in [trivial,simple,method]] cmd=['python3',os.path.expanduser('~/dev/ai-research-tools/tools/sbs/sbs.py'), '--trivial',str(errors[0]),'--simple',str(errors[1]),'--method',str(errors[2]), '--lower-is-better','--metric','MAE'] result=subprocess.run(cmd,capture_output=True,text=True,check=True) baselines[name]=json.loads(result.stdout) out['scores'][name]={'observed':obs.tolist(), 'MAE_percentage_points':dict(zip(['trivial','simple','method'],[100*x for x in errors])), 'SBS':(errors[0]-errors[1])/(errors[0]-errors[2])} corners=[] for signs in itertools.product([-1,1],repeat=4): obs=miba+np.array(signs)*np.array([.0005,.0005,.0005,.005]) err=[float(np.mean(abs(pred-obs))) for pred in [trivial,simple,method]] corners.append([*err,(err[0]-err[1])/(err[0]-err[2])]) out['rounding_bounds']={'min':np.min(corners,axis=0).tolist(),'max':np.max(corners,axis=0).tolist()} out['ssi_2021_integer_denominators']=[d for d in range(274,1000) if .6255<=274/d<.6265] out['same_year_source_contrasts_points']=((ssi-miba[:2])*100).tolist() Path('fresh_recomputation.json').write_text(json.dumps(out,indent=2)+'\n') Path('BASELINES.json').write_text(json.dumps(baselines,indent=2)+'\n') print(json.dumps(out,indent=2))