void GlobalGrid::makeBalanceWeights(){
int num_tensors = tensorList->getNumIndexes();
if ( tensor_weights != 0 ){ delete[] tensor_weights; }
tensor_weights = new int[ num_tensors ]; tzero( num_tensors, tensor_weights );
int max_level = computeMaxLevel();
for( int l = max_level; l>=0; l-- ){
for( int i=0; i<num_tensors; i++ ){
const int *iList = tensorList->getIndexList(i);
if ( tsum( num_dimensions, iList ) == l ){ // set the current level
tensor_weights[i] = 1 - tensor_weights[i];
for( int j=0; j<num_tensors; j++ ){
if ( (i!=j) && (isSubset(tensorList->getIndexList(j), iList)) ){
tensor_weights[j] += tensor_weights[i];
}
}
}
}
}
}
void GlobalGrid::reset( int dimensions, int outputs, int depth, TypeDepth type, TypeOneDRule oned, const int *anisotropic_weights, const double *alpha_beta ){
clear();
ruleType = oned;
num_dimensions = dimensions;
num_outputs = outputs;
if ( (oned == rule_gaussgegenbauer) || (oned == rule_gausslaguerre) || (oned == rule_gausshermite) ){
alpha = alpha_beta[0];
}
if ( (oned == rule_gaussjacobi) ){
alpha = alpha_beta[0];
beta = alpha_beta[1];
}
if ( anisotropic_weights != 0 ){
anisotropic = new int[num_dimensions+1];
tcopy( num_dimensions, anisotropic_weights, anisotropic );
anisotropic[num_dimensions] = tsum( num_dimensions, anisotropic_weights );
}
if ( type == type_level ){
makeOnedRule( depth );
}else if ( type == type_basis ){
makeOnedRule( 3 );
makeOnedRule( getLevelNeededForBasis( depth ) );
}else{
makeOnedRule( getLevelNeededForHyperbolic( depth ) );
}
makeTensorList( depth, type );
makeTensorsArray();
makeBalanceWeights();
makePoints();
}
//-----------------------------------------------------------------
void model::obs_pdf(double err){
obs_err_pdf.clear();
obs_err_cdf.clear();
log_obs_err_cdf.clear();
obs_err_pdf.resize(x.size());
obs_err_cdf.resize(x.size());
log_obs_err_cdf.resize(x.size());
obs_err = err;
vector<double> wgts_y;
double fac = 1./(obs_err*sqrt(2*3.14159265359));
long zero_ind=99999999999, ct=0;
long num_ind=0;
double diff_val=0;
for(int i =0;i<(long)x.size();i++){
//cout<<x[i]<<endl;
if(x[i]-x[0] > 5*obs_err){
num_ind = i;
diff_val = x[i]-x[0];
break;
}
}
for(double xx= -diff_val;xx<diff_val+log(pow(10,step))*.98; xx+=log(pow(10,step))){
if(fabs(xx) < 1e-8) zero_ind=ct;
ct++;
wgts_y.push_back(fac*exp(-0.5*(xx*xx)/(obs_err*obs_err)));
}
//exit(10);
double pdf_tmp, cdf_tmp;
long xsz = (long) x.size();
long wgts_ysz = (long) wgts_y.size();
long wgts_wing = wgts_ysz - zero_ind-1;
if(wgts_ysz != wgts_wing*2+1){
cout<<"model::obs_pdf wgts is not the right dimension.. logic bug"<<endl;
cout<<wgts_ysz<<"\t"<<wgts_wing<<"\t"<<zero_ind<<endl;
exit(1);
}
double sum=0;
for(int i=0; i<(long)x.size();i++){
sum=0;
pdf_tmp=0;
cdf_tmp=0;
for(int j= max((long)0,i-wgts_wing); j< min(i+wgts_wing, xsz-1); j++){
//cout<<j-i+zero_ind<<"\t"<<j<<"\t"<<i<<"\t"<<zero_ind<<endl;
sum += wgts_y[j-i+zero_ind];
pdf_tmp+=wgts_y[j-i+zero_ind]*pdf[j];
cdf_tmp+=wgts_y[j-i+zero_ind]*cdf[j];
}
obs_err_pdf[i]=(pdf_tmp/sum);
obs_err_cdf[i]=(cdf_tmp/sum);
}
long last = (long)x.size()-1;
double sum2 =tsum(x, obs_err_pdf);
for(int i=0; i<(long)x.size();i++){
obs_err_pdf[i] /= sum2;
//have to do pdf first
obs_err_cdf[i] /= obs_err_cdf[last];
log_obs_err_cdf[i] = log(obs_err_cdf[i]);
}
}
//-----------------------------------------------------------------
vector<double> model::sfr_pdf(double sfr){
vector<double> out;
out.resize(x.size());
double expr =0;
//double buffer =0;
double nhat = f_c*length*pow(10., sfr)/mean_m;
double tmp_log_nhat;
//cout<<sfr<<"\t"<<nhat<<"\t"<<f_c<<"\t"<<length<<"\t"<<pow(10., sfr)<<"\t"<<mean_m<<endl;
if(nhat < 100){
//if nhat is less than 100 then we can just brute force 300 iterations
//to get the full distribution
//I assume at least 1 cluser in each galaxy looked at because if there
// wasn't a cluster it probably wasn't included in this study
// THis is why the for loop starts at 1.... I will thus have to renormalize
tmp_log_nhat = log(nhat);
double jm1, tmp_fac;
for(double j=1; j<300;j+=1){
// j=floor(jj + 0.5); //round
jm1=j-1;
tmp_fac= j*exp( tmp_log_nhat*j- lgamma(j+1)-nhat);
for(int i=0;i<(long)x.size(); i++){
expr = exp( log_obs_err_cdf[i]*(jm1))*obs_err_pdf[i]*
tmp_fac;
if(expr != expr) expr=0;
out[i]+=expr;
}
}
//case of no clusters
cout<<exp(-nhat)<<endl;
out[0] += exp(-nhat);
/* for(int i=0;i<(long)x.size(); i++){
buffer = 0;
// tmp_log_obs_err_cdf = log(obs_err_cdf[i]);
tmp_log_nhat = log(nhat);
for(int j=1;j < 300; j++){
expr = j*exp( log_obs_err_cdf[i]*(j-1))*obs_err_pdf[i]*
exp( tmp_log_nhat*j- lgamma(j+1)-nhat);
if(expr != expr) expr=0;
buffer += expr;
}
out[i]=buffer;
}*/
} else {
//majority of the code lives in this loop!!!!
double sigma=sqrt(nhat);
double sig_step=sigma/10.;
double j=0;
double tmp_log_nhat, tmp_fac, jm1;
tmp_log_nhat = log(nhat);
for(double jj=nhat-5*sigma; jj<( nhat+5*sigma);jj+=sig_step){
j=floor(jj + 0.5); //round
jm1=j-1;
tmp_fac= j*exp( tmp_log_nhat*j- lgamma(j+1)-nhat);
for(int i=0;i<(long)x.size(); i++){
expr = exp( log_obs_err_cdf[i]*(jm1))*obs_err_pdf[i]*
tmp_fac;
if(expr != expr) expr=0;
out[i]+=expr;
}
}
}
double sum = tsum(x, out);
for(int i = 0;i<(long)x.size();i++) out[i] /= sum;
//cout<<out.size()<<endl;
return(out);
}
int GlobalGrid::computeMaxLevel() const{
int max_level = 0;
for( int i=0; i<tensorList->getNumIndexes(); i++ ){
int sum = tsum( num_dimensions, tensorList->getIndexList(i) );
max_level = ( sum > max_level ) ? sum : max_level;
}
return max_level;
}