static double aterms_mu(double mya, void *mydata) {
int e, t;
double val = 0;
#ifdef A_DEBUG
float save_a = ddC.a_mu->a;
double like;
#endif
S_remake(ddC.a_mu, mya);
for (e=0; e<ddN.E; e++) {
for (t=0; t<ddN.T; t++) {
if ( ddS.cp_et[e][t]==0 )
continue;
if (e==ddN.E-1)
val += S_S(ddC.a_mu, ddS.C_eDt[e][t], ddS.cp_et[e][t]);
else
val +=
S_S(ddC.a_mu, ddS.C_eDt[e][t] + ddS.cp_et[e+1][t], ddS.cp_et[e][t]);
}
val += poch(ddP.b_mu[e], mya, ddS.Cp_e[e]);
}
myarms_evals++;
#ifdef A_DEBUG
yap_message("Eval aterms_mu(%lf) = %lf (S had %f)", mya, val, save_a);
ddP.a_mu = mya;
cache_update("am");
like = likelihood();
if ( last_val != 0 ) {
yap_message(", lp=%lf diffs=%lf vs %lf\n", like,
val-last_val, like-last_like);
}
last_like = like;
last_val = val;
#endif
return val;
}
double likelihood_PYalpha() {
int i,t;
double likelihood = 0;
double la = 0;
double lb = log(ddP.bpar);
if ( ddP.apar>0 ) la = log(ddP.apar);
for (i=0; i<ddN.DT; i++) {
uint16_t Td_ = 0;
for (t=0; t<ddN.T; t++) {
if ( ddS.Ndt[i][t]>0 ) {
Td_ += ddS.Tdt[i][t];
if ( ddS.Ndt[i][t]>1 ) {
likelihood += S_S(ddC.SX,ddS.Ndt[i][t],ddS.Tdt[i][t]);
}
}
}
if ( ddP.apar==0 ) {
likelihood += Td_*lb;
} else {
likelihood += Td_*la + gcache_value(&ddC.lgba, (int)Td_);
}
likelihood -= gcache_value(&ddC.lgb, (int)ddS.NdT[i]);
//yap_infinite(likelihood);
}
return likelihood;
}
static double aterms_phi0(double mya, void *mydata) {
int v;
double val = 0;
#ifdef A_DEBUG
float save_a = ddC.a_phi0->a;
double like;
#endif
S_remake(ddC.a_phi0, mya);
val += poch(ddP.b_phi0, mya, ddS.S_0_nz);
for (v=0; v<ddN.W; v++) {
if ( ddS.S_0vT[v]>1 )
val += S_S(ddC.a_phi0, ddS.S_0vT[v], 1);
}
myarms_evals++;
#ifdef A_DEBUG
yap_message("Eval aterms_phi0(%lf) = %lf (S had %f)", mya, val, save_a);
ddP.a_phi0 = mya;
cache_update("ap0");
like = likelihood();
if ( last_val != 0 ) {
yap_message(", lp=%lf diffs=%lf vs %lf\n", like,
val-last_val, like-last_like);
}
last_like = like;
last_val = val;
#endif
return val;
}
static double aterms_theta(double mya, void *mydata) {
int i, t;
double val = 0;
#ifdef A_DEBUG
float save_a = ddC.a_theta->a;
double like;
#endif
S_remake(ddC.a_theta, mya);
for (i=0; i<ddN.DT; i++) {
for (t=0; t<ddN.T; t++) {
if ( ddS.n_dt[i][t]>1 ) {
val += S_S(ddC.a_theta,ddS.n_dt[i][t],ddS.c_dt[i][t]);
}
}
val += poch(ddP.b_theta, mya, ddS.C_dT[i]);
}
myarms_evals++;
#ifdef A_DEBUG
yap_message("Eval aterms_theta(%lf) = %lf (S had %f)", mya, val, save_a);
ddP.a_theta = mya;
cache_update("at");
like = likelihood();
if ( last_val != 0 ) {
yap_message(", lp=%lf diffs=%lf vs %lf\n", like,
val-last_val, like-last_like);
}
last_like = like;
last_val = val;
#endif
return val;
}
static double aterms(double mya, void *mydata) {
int i, t;
double val = 0;
double la = log(mya);
#ifdef A_DEBUG
float save_a = ddC.SX->a;
double like;
#endif
S_remake(ddC.SX, mya);
for (i=0; i<ddN.DT; i++) {
uint32_t Td_ = 0;
for (t=0; t<ddN.T; t++) {
Td_ += ddS.Tdt[i][t];
if ( ddS.Ndt[i][t]>1 ) {
val += S_S(ddC.SX,ddS.Ndt[i][t],ddS.Tdt[i][t]);
}
}
val += Td_*la + lgamma(ddP.bpar/mya+Td_) - lgamma(ddP.bpar/mya);
}
myarms_evals++;
#ifdef A_DEBUG
yap_message("Eval aterms(%lf) = %lf (S had %f)", mya, val, save_a);
ddP.apar = mya;
cache_update("a");
like = likelihood();
if ( last_val != 0 ) {
yap_message(", lp=%lf diffs=%lf vs %lf\n", like,
val-last_val, like-last_like);
}
last_like = like;
last_val = val;
#endif
return val;
}
static double awterms(double myaw, void *mydata) {
int i, t;
double val = 0;
double law = log(myaw);
#ifdef A_DEBUG
float save_a = ddC.SY->a;
double like;
#endif
S_remake(ddC.SY, myaw);
for (t=0; t<ddN.T; t++) {
uint32_t Tw_ = 0;
for (i=0; i<ddN.W; i++) {
Tw_ += ddS.Twt[i][t];
if ( ddS.Nwt[i][t]>1 ) {
val += S_S(ddC.SY,ddS.Nwt[i][t],ddS.Twt[i][t]);
}
}
val += Tw_*law + lgamma(ddP_bwpar(t)/myaw+Tw_) - lgamma(ddP_bwpar(t)/myaw);
}
myarms_evals++;
#ifdef A_DEBUG
yap_message("Eval awterms(%lf) = %lf (S had %f)", myaw, val, save_a);
ddP.awpar = myaw;
cache_update("aw");
like = likelihood();
if ( last_val != 0 ) {
yap_message(", lp=%lf diffs=%lf vs %lf\n", like,
val-last_val, like-last_like);
}
last_like = like;
last_val = val;
#endif
return val;
}
static double aterms_phi1(double mya, void *mydata) {
int e, t, v;
double val = 0;
#ifdef A_DEBUG
float save_a = ddC.a_phi1->a;
double like;
#endif
S_remake(ddC.a_phi1, mya);
for (e=0; e<ddN.E; e++) {
for (t=0; t<ddN.T; t++) {
if ( ddS.S_Vte[t][e]==0 )
continue;
val += poch(ddP.b_phi[e][t], mya, ddS.S_Vte[t][e]);
for (v=0; v<ddN.W; v++) {
if ( ddS.s_vte[v][t][e]==0 )
continue;
if (e<ddN.E-1) {
val += S_S(ddC.a_phi1, ddS.m_vte[v][t][e] + ddS.s_vte[v][t][e+1] , ddS.s_vte[v][t][e]);
} else {
val += S_S(ddC.a_phi1, ddS.m_vte[v][t][e], ddS.s_vte[v][t][e]);
}
}
}
}
myarms_evals++;
#ifdef A_DEBUG
yap_message("Eval aterms_phi1(%lf) = %lf (S had %f)", mya, val, save_a);
ddP.a_phi1 = mya;
cache_update("ap"1);
like = likelihood();
if ( last_val != 0 ) {
yap_message(", lp=%lf diffs=%lf vs %lf\n", like,
val-last_val, like-last_like);
}
last_like = like;
last_val = val;
#endif
return val;
}
static double aterms(double x, void *mydata) {
int i, k;
ALData *mp = (ALData *)mydata;
double val = 0;
if ( x<=0 ) {
fprintf(stderr,"Illegal discount value in aterms()\n");
exit(1);
}
if ( mp->verbose>1 ) {
fprintf(stderr,"Extending S for M=%d a=%lf\n", mp->maxt,x);
}
if ( mp->S )
S_remake(mp->S,x);
else
mp->S = S_make(mp->maxn,mp->maxt,mp->maxn,mp->maxt,x,S_STABLE);
if ( !mp->S ) {
fprintf(stderr,"Out of memory for S table\n");
exit(1);
}
for (i=0; i<mp->I; i++) {
val += mp->T[i] * log(x) +
lgamma(mp->T[i]+mp->bpar[i]/x) - lgamma(mp->bpar[i]/x);
if ( mp->val ) {
for (k=0; k<mp->K[i]; k++) {
scnt_int n;
stcnt_int t;
mp->val(&n, &t, i, k);
if ( n>1 )
val += S_S(mp->S, n, t);
}
} else {
for (k=0; k<mp->K[i]; k++)
if ( mp->n[i][k]>1 )
val += S_S(mp->S, mp->n[i][k],mp->t[i][k]);
}
}
return val;
}
int main(int argc, char* argv[])
{
int c;
int n, t;
stable_t *S;
/*
* default values for args
*/
while ( (c=getopt(argc, argv,"ha:N:T:"))>=0 ) {
switch ( c ) {
case 'a':
if ( !optarg || sscanf(optarg,"%f",&apar)!=1 )
yaps_quit("Need a valid 'a' argument\n");
break;
case 'N':
if ( !optarg || sscanf(optarg,"%d",&N)!=1 )
yaps_quit("Need a valid 'N' argument\n");
break;
case 'T':
if ( !optarg || sscanf(optarg,"%d",&T)!=1 )
yaps_quit("Need a valid 'T' argument\n");
break;
case 'h':
usage();
exit(0);
}
}
if ( T>N ) T = N;
if ( !(S=S_make(N+1,T,2*N,2*T,apar,S_UVTABLE|S_STABLE|S_VERBOSE)) )
yaps_quit("S_make failed\n");
S_report(S, stdout);
#if 1
/*
* list various values
*/
printf("S(%d,%d) = %10.6lg, V=n/a U=%lg\n", N, 1, S_S(S,N,1), S_U(S,N,1));
for (t=2; t<=T; t++)
printf("S(%d,%d) = %10.6lg, V=%lg U=%lg\n", N, t,
S_S(S,N,t), S_V(S,N,t), S_U(S,N,t));
printf("\nS(%d,%d) = %10.6lg, V=n/a U=%lg\n",
N+10, 1, S_S(S,N+10,1), S_U(S,N+10,1));
for (t=2; t<=2*T+10; t++)
printf("S(%d,%d) = %10.6lg, V=%lg U=%lg\n", N+10, t,
S_S(S,N+10,t), S_V(S,N+10,t), S_U(S,N+10,t));
for (n=T+1; n<=2*N; n++)
printf("S(%d,%d..) = %10.6lg %10.6lg\n", n, T, S_S(S,n,T), S_S(S,n,T+1));
for (n=2; n<8; n++ ) {
printf("S(%d,%d) = %10.6lg ", n, 1, S_S(S,n,1));
for (t=2; t<=n; t++)
printf(" %10.6lg", S_S(S,n,t));
printf("\n");
}
for (n=2; n<8; n++ ) {
printf("V(%d,%d) = %10.6lg", n, 2, S_V(S,n,2));
for (t=3; t<=n; t++)
printf(" %10.6lg", S_V(S,n,t));
printf("\n");
}
for (n=2; n<8; n++ ) {
printf("U(%d,%d) = %10.6lg", n, 2, S_U(S,n,2));
for (t=3; t<=n; t++)
printf(" %10.6lg", S_U(S,n,t));
printf("\n");
}
#else
{
/*
* Sample a partition of size T of N by Chinese Rest. distribution
* start by sampling the last entry from (1,2,...,N-T+1);
* see p(m | CRD, apr, N, T) on page 4 of "doc/alpha.pdf"
*/
double *prob = malloc(sizeof(*prob)*N);
double probtot = 1.0;
prob[1] = 1.0;
for (t=2; t<=N-T+1; t++)
probtot += prob[t] = (N-t+1)*(t-apar) / S_U(S,N-t,T-1)/(t-1) * prob[t-1];
for (t=1; t<=N-T+1; t++)
printf("p(m=%d) = %lg\n", t, prob[t]/probtot );
}
#endif
S_report(S, stdout);
S_free(S);
return 0;
}
double likelihood_PYbeta() {
int i,t;
double likelihood = 0;
double lbw = log(ddP.bwpar);
double law = log(ddP.awpar);
likelihood += pctl_gammaprior(ddP.bwpar);
/*
* term for k-th node
*/
#ifdef BWPAR0
for (t=1; t<ddN.T; t++) {
#else
for (t=0; t<ddN.T; t++) {
#endif
uint32_t Tw_ = 0;
for (i=0; i<ddN.W; i++) {
int tt = ddS.Twt[i][t];
int nn = ddS.Nwt[i][t];
if ( nn>0 ) {
Tw_ += tt;
likelihood += S_S(ddC.SY,nn,tt);
#if 1
if ( !finite(likelihood) || isinf(likelihood) || isnan(likelihood) )
yap_quit("Like=%lf: Nwt[%d][%d]=%d Twt[i][t]=%d S.M=%d S.N=%d\n",
likelihood,
i, t, (int)ddS.Nwt[i][t],(int)ddS.Twt[i][t],ddC.SY->usedM, ddC.SY->usedN);
#endif
}
}
yap_infinite(likelihood);
if ( ddP.awpar==0 ) {
likelihood += Tw_*lbw;
} else {
#ifdef L_CACHE
likelihood += Tw_*law + gcache_value(&ddC.lgbaw, (int)Tw_);
#else
likelihood += Tw_*law + gammadiff((int)Tw_, ddP.bwpar/ddP.awpar, 0.0);
#endif
}
#ifdef L_CACHE
likelihood -= gcache_value(&ddC.lgbw, (int)ddS.NWt[t]);
#else
likelihood -= gammadiff((int)ddS.NWt[t], ddP.bwpar, 0.0);
#endif
yap_infinite(likelihood);
}
yap_infinite(likelihood);
return likelihood;
}
double likelihood_PYbeta_PDP() {
/*
* the constant prior
*/
int j;
double likelihood = 0;
for (j=0; j<ddN.W; j++) {
if ( ddS.TwT[j]>0 ) {
likelihood += ddS.TwT[j]*log(ddP.betapr[j]);
}
}
//yap_infinite(likelihood);
return likelihood;
}
/*
* use prior a is uniform
*/
double samplea2(double mya, stable_t *S,
int I, int *K, scnt_int *T,
scnt_int **n, stcnt_int **t,
void (*getval)(scnt_int *n, stcnt_int *t, unsigned i, unsigned k),
double *bpar,
rngp_t rng, int loops, int verbose) {
double inita[3] = {A_MIN,1,A_MAX};
int i, k;
ALData ald;
stcnt_int *mp;
int n_m = 0;
inita[1] = mya;
if ( fabs(inita[1]-A_MAX)/A_MAX<0.00001 ) {
inita[1] = A_MAX*0.999 + A_MIN*0.001;
}
if ( fabs(inita[1]-A_MIN)/A_MIN<0.00001 ) {
inita[1] = A_MIN*0.999 + A_MAX*0.001;
}
#ifdef SQUEEZEA
/*
* bound current move to be less than SQEEZEA
*/
if ( inita[1]-SQUEEZEA>A_MIN ) inita[0] = inita[1]-SQUEEZEA;
if ( inita[1]+SQUEEZEA<A_MAX ) inita[2] = inita[1]+SQUEEZEA;
#endif
ald.T = T;
ald.n = n;
ald.t = t;
ald.I = I;
ald.K = K;
ald.val = getval;
ald.bpar = bpar;
ald.verbose = verbose;
ald.maxt = 1;
ald.maxn = 1;
ald.S = NULL;
n_m = 0;
for (i=0; i<I; i++)
for (k=0; k<K[i]; k++) {
if ( t[i][k]>1 && t[i][k]<n[i][k])
n_m += t[i][k]-1 ;
}
ald.m = malloc(sizeof(*ald.m)*n_m);
if ( !ald.m ) {
fprintf(stderr,"Out of memory for samplea()\n");
exit(1);
}
// fprintf(stderr,"m[%d][%d] tabl space = %d\n", I, K[0], n_m);
mp = ald.m;
for (i=0; i<I; i++)
for (k=0; k<K[i]; k++) {
if ( t[i][k]>1 && t[i][k]<n[i][k]) {
int N = n[i][k];
double ptot = S_S(S,N,t[i][k]);
double rem = ptot + log(rng_unit(rng));
int M;
// fprintf(stderr,"Sampling[%d][%d] n=%d, t=%d: ", i,k, N, t[i][k]);
for (M=t[i][k]-1; M>=1; M--) {
// each round instantiates another count
int l;
double fact = 0.0;
for (l=1; l<=N-M; l++) {
double term;
if ( l>1 )
fact += log((l-mya)*(N-l+1)/(l-1));
term = fact+S_S(S,N-l,M)-ptot;
if ( term>= rem )
break;
rem = logminus(rem, term);
}
if ( l>N-M ) l = N-M;
mp[M-1] = l;
N -= l;
assert(N>=1);
// fprintf(stderr," %d", l);
}
// fprintf(stderr," %d\n", N);
mp += t[i][k]-1;
}
}
#ifdef PSAMPLE_ARS
arms_simple(3, inita, inita+2, aterms2,
NULL, 0, inita+1, &mya);
if ( mya<inita[0] || mya>inita[2] ) {
fprintf(stderr,"Arms_simple(apar) returned value out of bounds\n");
exit(1);
}
#else
inita[1] = A_MAX;
if ( SliceSimple(&mya, aterms2, inita, rng, loops, &ald) ) {
fprintf(stderr,"SliceSimple error\n");
exit(1);
}
#endif
if ( ald.S ) S_free(ald.S);
return mya;
}
/*
* compute likelihood ratio difference based on *M
*/
static double merge_like_Tdt_sum(int k1, int k2, merge_alpha_t *M) {
double *val;
int d;
val = dvec(ddN.DT);
for (d=0; d<ddN.DT; d++) {
}
#endif
static double merge_like_Tdt(int k1, int k2, merge_alpha_t *M) {
int d;
double la = 0;
double lb = log(ddP.bpar);
int TD_diff = 0;
double likelihood = 0;
if ( ddP.apar>0 ) la = log(ddP.apar);
for (d=0; d<ddN.DT; d++) {
int Td_diff; /* total change in T for doc */
if ( M->Ndt[d]>1 ) {
likelihood -= S_S(ddC.SX,ddS.Ndt[d][k2],ddS.Tdt[d][k2]);
likelihood -= S_S(ddC.SX,ddS.Ndt[d][k1],ddS.Tdt[d][k1]);
likelihood += S_S(ddC.SX,M->Ndt[d],M->Tdt[d]);
assert(M->Tdt[d]>=1);
assert(M->Tdt[d]<=M->Ndt[d]);
assert(ddS.Ndt[d][k2]==0 || ddS.Tdt[d][k2]>0);
assert(ddS.Ndt[d][k1]==0 || ddS.Tdt[d][k1]>0);
}
yap_infinite(likelihood);
TD_diff += Td_diff = (M->Tdt[d]-ddS.Tdt[d][k2]-ddS.Tdt[d][k1]);
if ( Td_diff==0 )
continue;
if ( ddP.apar==0 ) {
likelihood += Td_diff*lb;
} else {
likelihood += Td_diff*la;
if ( Td_diff<0 )
likelihood -=
gammadiff(-Td_diff,M->TdT[d]+ddP.bpar/ddP.apar, 0.0);
else
likelihood +=
gammadiff(Td_diff, M->TdT[d]-Td_diff+ddP.bpar/ddP.apar, 0.0);
}
yap_infinite(likelihood);
}
if ( ddP.PYalpha==H_PDP ) {
likelihood += (M->TDt-ddS.TDt[k1])*log(ddP.alphapr[k1])
- ddS.TDt[k2]*log(ddP.alphapr[k2]);
} else if ( ddP.PYalpha==H_HDP ) {
likelihood += lgamma(M->TDTm+M->TDt-TD_diff+ddP.b0)
- lgamma(M->TDTm+M->TDt+ddP.b0);
likelihood -= gammadiff(ddS.TDt[k1], ddP.b0*ddP.alphapr[k1], 0.0);
likelihood -= gammadiff(ddS.TDt[k2], ddP.b0*ddP.alphapr[k2], 0.0);
likelihood += gammadiff(M->TDt, ddP.b0*ddP.alphapr[k1], 0.0);
} else {
double lga0 = lgamma(1-ddP.a0);
likelihood += lgamma(M->TDTm+M->TDt-TD_diff+ddP.b0)
- lgamma(M->TDTm+M->TDt+ddP.b0);
/* because k2 gone to zero, so one less topic */
likelihood -= log(ddP.b0+ddP.a0*(ddS.TDTnz-1));
if ( ddS.TDt[k2]>1 )
likelihood -= lgamma(ddS.TDt[k2]-ddP.a0) - lga0;
if ( ddS.TDt[k1]>1 )
likelihood -= lgamma(ddS.TDt[k1]-ddP.a0) - lga0;
likelihood += lgamma(M->TDt-ddP.a0) - lga0;
}
yap_infinite(likelihood);
return likelihood;
}
/*
* compute likelihood ratio difference based on *M
*/
static double merge_like_Twt(int k1, int k2, merge_beta_t *M) {
int i, w;
double likelihood = 0;
#ifndef BWPAR0
double lbw = log(ddP.bwpar);
#endif
double law = log(ddP.awpar);
double TW_diff = 0;
#ifdef BWPAR0
yap_quit("BWPAR0 unimpleented in merge\n");
#endif
for (i=0; i<ddN.W; i++) {
likelihood -= S_S(ddC.SY,ddS.Nwt[i][k1],ddS.Twt[i][k2]);
likelihood -= S_S(ddC.SY,ddS.Nwt[i][k1],ddS.Twt[i][k2]);
likelihood += S_S(ddC.SY,M->Nwt[i],M->Twt[i]);
}
if ( ddP.awpar==0 ) {
#ifdef BWPAR0
likelihood += M->TWt-ddS.TWt[k1]*log(ddP_bwpar(k1))
-ddS.TWt[k2]*log(ddP_bwpar(k2));
#else
likelihood += (M->TWt-ddS.TWt[k1]-ddS.TWt[k2])*lbw;
#endif
} else {
likelihood += (M->TWt-ddS.TWt[k1]-ddS.TWt[k2])*law
+ gammadiff((int)M->TWt, ddP.bwpar/ddP.awpar, 0.0)
- gammadiff((int)ddS.TWt[k1], ddP_bwpar(k1)/ddP.awpar, 0.0)
- gammadiff((int)ddS.TWt[k2], ddP_bwpar(k2)/ddP.awpar, 0.0);
}
likelihood += gammadiff((int)ddS.NWt[k1], ddP_bwpar(k1), 0.0);
likelihood += gammadiff((int)ddS.NWt[k2], ddP_bwpar(k2), 0.0);
likelihood -= gammadiff((int)M->NWt, ddP.bwpar, 0.0);
yap_infinite(likelihood);
if ( ddP.PYbeta==H_PDP ) {
for (w=0; w<ddN.W; w++) {
if ( ddS.TwT[w]>0 ) {
// ???????????????
likelihood += ddS.TwT[w]*log(ddP.betapr[w]);
}
}
} else if ( ddP.PYbeta==H_HDP ) {
yap_quit("merge with PYbeta unimplemented\n");
likelihood += lgamma(M->TWTm+M->TWt-TW_diff+ddP.bw0)
- lgamma(M->TWTm+M->TWt+ddP.bw0);
for (w=0; w<ddN.W; w++) {
// ???????????
likelihood -= gammadiff(ddS.TWt[k1], ddP.bw0*ddP.betapr[k1], 0.0);
likelihood -= gammadiff(ddS.TWt[k2], ddP.bw0*ddP.betapr[k2], 0.0);
likelihood += gammadiff(M->TWt, ddP.bw0*ddP.betapr[k1], 0.0);
}
} else {
double lgaw0 = lgamma(1-ddP.aw0);
likelihood += lgamma(M->TWTm+M->TWt-TW_diff+ddP.bw0)
- lgamma(M->TWTm+M->TWt+ddP.bw0);
/* because k2 gone to zero, so one less topic */
likelihood -= log(ddP.bw0+ddP.aw0*(ddS.TWTnz-1));
if ( ddS.TWt[k2]>1 )
likelihood -= lgamma(ddS.TWt[k2]-ddP.aw0) - lgaw0;
if ( ddS.TWt[k1]>1 )
likelihood -= lgamma(ddS.TWt[k1]-ddP.aw0) - lgaw0;
likelihood += lgamma(M->TWt-ddP.aw0) - lgaw0;
}
yap_infinite(likelihood);
return likelihood;
}