# E-M1b impl2 · migración (§3) y Carril (§5). numpy/pandas/statsmodels (imagen bquantagents-worker). Código propio. import numpy as np, pandas as pd, statsmodels, statsmodels.api as sm, sys, math, json, time TMP='/w/tmp/impl2'; OUT='/w/salidas/impl2' t0=time.time() ver=f"python {sys.version.split()[0]} · numpy {np.__version__} · pandas {pd.__version__} · statsmodels {statsmodels.__version__}" print(ver) d=pd.read_csv(f'{TMP}/lic.csv', parse_dates=['fecha_ref']) for c in ['age','es_na','tiene_cpv','obj_ok']: d[c]=d[c].astype(str).str.lower().eq('true') d['ss']=d.tipo_contrato.isin(['Servicios','Suministros']) d['obras']=d.tipo_contrato.eq('Obras') d['oc']=d.tipo_contrato.isin(['Obras','Concesion_obras','Concesion_servicios']) print('filas', len(d), d.feed.value_counts().to_dict()) def periodo(f, a, b): # trimestres inclusive 'YYYYTq' return (d.trimestre>=a)&(d.trimestre<=b)&(d.feed==f) # --------- (i) migración MIG=[] celdas_m={ 'G1':dict(pob=lambda: d.ss & d.age, clase='ss', c26=(143000,140000), c24=(140000,143000)), 'G2':dict(pob=lambda: d.ss & ~d.age & ~d.es_na, clase='ss', c26=(143000,140000), c24=(140000,143000)), 'G3':dict(pob=lambda: d.ss & ~d.age, clase='ss', c26=(221000,216000), c24=(215000,221000)), 'G4':dict(pob=lambda: d.oc, clase='oc', c26=(5538000,5404000), c24=(5382000,5538000)), } PER={'2026':[('antes','2024T1','2025T4'),('despues2T','2026T1','2026T2'),('despues','2026T1','2026T3')], '2024':[('antes','2022T1','2023T4'),('despues2T','2024T1','2024T2'),('despues','2024T1','2025T4')]} def migracion(celda, feed, rep): c=celdas_m[celda]; Tv,Tn=c['c26'] if rep=='2026' else c['c24']; w=abs(Tv-Tn) ve=d.ve if TnTn)&(ve=Tn-w)&(ve=Tn-3*w)&(ve=Tv-w)&(veTv)&(ve=Tv-4*w)&(ve=150000)&(ve<200000); unidad=1000 else: norm=(ve>=3200000)&(ve<5000000); unidad=100000 pob=c['pob']() rows=[] for nom,a,b in PER[rep]: m=pob & periodo(feed,a,b) nv,nn,nr,nz=int((m&vieja).sum()),int((m&nueva).sum()),int((m&ref).sum()),int((m&norm).sum()) dens=lambda n,anch: (1000*n/nz)/(anch/unidad) if nz>0 else float('nan') rows.append(dict(celda=celda+('' if rep=='2026' else '_r2024'), feed=feed, periodo=nom, n_vieja=nv,n_nueva=nn,n_ref=nr,n_norm=nz,d_vieja=dens(nv,w),d_nueva=dens(nn,w),d_ref=dens(nr,3*w))) A,D2,D=rows ev = A['n_norm']>=200 and D2['n_norm']>=100 and A['n_vieja']>=20 and A['d_vieja']>=2*A['d_ref'] for r in rows: r['evaluable']=ev if r['periodo']=='antes': r['migracion']=''; r['caida']=float('nan') else: r['migracion']= (r['d_vieja']<=0.5*A['d_vieja']) if ev else ('sin_potencia' if not (A['n_norm']>=200 and D2['n_norm']>=100 and A['n_vieja']>=20) else 'no_pico') r['caida']=1-r['d_vieja']/A['d_vieja'] if A['d_vieja']>0 else float('nan') r['subida_nueva']=(r['d_nueva']/A['d_nueva']) if A['d_nueva']>0 else float('nan') return rows for rep in ['2026','2024']: for g in ['G1','G2','G3','G4']: MIG+=migracion(g,'propio',rep) MIG+=migracion('G3','agregacion',rep) mig=pd.DataFrame(MIG) # serie trimestral (potencia) SER=[] for g in ['G1','G2','G3','G4']: for feed in (['propio','agregacion'] if g=='G3' else ['propio']): c=celdas_m[g]; pob=c['pob']()&(d.feed==feed) for rep,(Tv,Tn) in [('2026',c['c26']),('2024',c['c24'])]: w=abs(Tv-Tn); ve=d.ve if TnTn)&(ve=Tn-w)&(ve=Tv-w)&(veTv)&(ve=150000)&(ve<200000)) if c['clase']=='ss' else ((ve>=3200000)&(ve<5000000)) t=pd.DataFrame({'q':d.trimestre[pob],'v':vieja[pob],'n':nueva[pob],'z':norm[pob]}).groupby('q').sum() for q,r in t.iterrows(): SER.append(dict(celda=g,feed=feed,cambio=rep,trimestre=q,n_vieja=int(r.v),n_nueva=int(r.n),n_norm=int(r.z))) pd.DataFrame(SER).to_csv(f'{OUT}/impl2_migracion_serie.csv',index=False) cols=['celda','feed','periodo','n_vieja','n_nueva','n_ref','n_norm','d_vieja','d_nueva','d_ref','evaluable','migracion','caida'] mig[cols].to_csv(f'{OUT}/impl2_migracion.csv',index=False,float_format='%.6g') mig.to_csv(f'{OUT}/impl2_migracion_ext.csv',index=False,float_format='%.6g') # --------- (ii) Carril def entre(a,b): return (d.fecha_ref>=pd.Timestamp(a))&(d.fecha_ref<=pd.Timestamp(b)) CEL=[ # nombre, pob, feed, periodo, T, estricto, w, lo, hi, vlo, R1, R2, s ('C1', lambda: d.ss&d.age, 'propio', ('2024-01-01','2025-12-31'), 143000, True, 1000, 105000, 143000, 140000, 5000, 10000, 10000), ('C1r',lambda: d.ss&d.age, 'propio', ('2026-01-01','2026-09-30'), 140000, True, 1000, 105000, 140000, 137000, 5000, 10000, 10000), ('C2', lambda: d.ss&~d.age&~d.es_na, 'propio', ('2024-01-01','2025-12-31'), 143000, True, 1000, 105000, 143000, 140000, 5000, 10000, 10000), ('C3', lambda: d.ss&~d.age, 'propio', ('2024-01-01','2025-12-31'), 221000, True, 1000, 150000, 221000, 216000, 5000, 10000, 10000), ('C3b',lambda: d.ss&~d.age, 'agregacion', ('2024-01-01','2025-12-31'), 221000, True, 1000, 150000, 221000, 216000, 5000, 10000, 10000), ('C4', lambda: d.ss&~d.es_na, 'propio', ('2024-01-01','2026-09-30'), 100000, False, 500, 65000, 100000, 97000, 5000, 10000, 10000), ('C5', lambda: d.obras&~d.es_na, 'propio', ('2024-01-01','2026-09-30'), 2000000, False, 25000, 1050000, 2000000, 1900000, 50000, 100000, 100000), ('C6', lambda: d.oc&~d.es_na, 'propio', ('2024-01-01','2026-09-30'), 3000000, False, 50000, 2100000, 3000000, 2850000, 100000, 500000, 100000), ('C7', lambda: d.oc, 'propio', ('2024-01-01','2025-12-31'), 5538000, True, 100000, 3238000, 5538000, 5338000, 500000, 1000000, 1000000), ] def contiene_multiplo(a,b,R,estricto): # estricto: [a,b) ; inclusivo: (a,b] if estricto: m=math.ceil(a/R)*R; return ma def disenio(T,estricto,w,lo,hi,vlo,R1,R2,s): K=int(round((hi-lo)/w)); assert abs((hi-lo)/w-K)<1e-9 kw=int(round((T-vlo)/w)) rows=[] for k in range(K): a=T-(k+1)*w; b=T-k*w rows.append(dict(k=k,a=a,b=b,z=((a+b)/2-T)/s,d1=int(contiene_multiplo(a,b,R1,estricto)),d2=int(contiene_multiplo(a,b,R2,estricto)),win=k=lo)&(velo)&(ve<=hi) sub=d[m][inr.values].copy() sub['k']=bin_k(sub.ve.values,T,estr,w) ds=disenio(T,estr,w,lo,hi,vlo,R1,R2,s); K=len(ds) assert sub.k.min()>=0 and sub.k.max()=0.70 else ('troceo' if x<=0.40 else 'intermedia') if row['n_a']<2000 or row['n_b']<500 or row['n_c']<500: res='muestra insuficiente' elif not (row['b_a_lo']>0 or row['b_a_hi']<0): res='sin amontonamiento que explicar' elif row['b_a']<0: res='sin amontonamiento que explicar (déficit)' else: zb,zc=zona(row['rho_b']),zona(row['rho_c']) if zb==zc=='recorte': v='ajuste del importe (recorte)' elif zb==zc=='troceo': v='fraccionamiento (troceo)' else: v='mixto' + (' discordante' if {zb,zc}=={'recorte','troceo'} else '') fb=zona(row['rho_b_lo'])==zona(row['rho_b_hi'])==zb fc=zona(row['rho_c_lo'])==zona(row['rho_c_hi'])==zc res=v+(' · firme' if (fb and fc) else ' · en el límite') row['resultado']=res CAR.append(row) ds['n_a']=H['a'].sum(0); ds['n_b']=H['b'].sum(0); ds['n_c']=H['c'].sum(0); ds['celda']=nom DET[nom]=ds print(nom, res, round(time.time()-t0,1),'s', flush=True) car=pd.DataFrame(CAR) cc=['celda','n_a','n_b','n_c','nv_a','nv_b','nv_c','org_boot','b_a','b_a_lo','b_a_hi','b_b','b_b_lo','b_b_hi','b_c','b_c_lo','b_c_hi','rho_b','rho_b_lo','rho_b_hi','rho_c','rho_c_lo','rho_c_hi','resultado'] car[cc].to_csv(f'{OUT}/impl2_carril.csv',index=False,float_format='%.6g') car.to_csv(f'{OUT}/impl2_carril_ext.csv',index=False,float_format='%.6g') pd.concat(DET.values()).to_csv(f'{OUT}/impl2_carril_bins.csv',index=False) open(f'{OUT}/versiones_estimacion.txt','w').write(ver+'\n') print('fin', round(time.time()-t0,1))