void MakeMarginals(LPTable Y,LPTable dataTable)
{
int i;
int lenC = dataTable->nDimens;
int* C = new int[lenC];
if(!Y->Alloc(dataTable->Dimens,dataTable->nDimens))
{
printf("Failed to allocate marginals table.\n"); exit(1);
}
Y->GetFirst();
Y->Set(dataTable->GetGrandTotal());
while(Y->GetNext())
{
lenC = 0;
for(i=0;i<Y->nDimens;i++)
{
if(Y->Index[i])
{
C[lenC] = i;
lenC++;
}
}
LPNTable p = new NTable;
p->Create(lenC,C,dataTable);
Y->Set(p->Data[p->Total-1]);
p->Reset(); delete p; p = NULL;
}
delete[] C; C = NULL;
return;
}
void PfromTheta(LPTable P,LPTable Theta)
{
int i;
double s;
double total;
total = 0.0;
P->GetFirst();
while(P->GetNext())
{
Theta->GetFirst();
s = 0.0;
while(Theta->GetNext())
{
if(subset(Theta->nDimens,Theta->Index,P->Index))
{
s += Theta->Get();
}
}
P->Set(exp(s));
total += P->Get();
}
P->Data[0] = 1.0/(1.0+total);
for(i=1;i<P->Total;i++)
{
P->Data[i] *= P->Data[0];
}
return;
}
LPTable CreateMin(int m, LPTable parent)
{
STable stab;
if(!stab.Create(m, parent))
{
printf("Error creating STable :: CreateS.\n");
return(NULL);
}
LPTable newtab = new Table;
if(newtab == NULL)
{
printf("Error creating STable :: CreateS.\n");
return(NULL);
}
if(!newtab->Alloc(parent->Dimens, parent->nDimens))
{
printf("Error creating STable :: CreateS.\n");
return(NULL);
}
int NotFinished = 1;
newtab->GetFirst();
while(NotFinished)
{
newtab->Set(stab.GetMin(newtab->Index));
NotFinished = newtab->GetNext();
}
stab.Reset();
return newtab;
}
LPTable Table::ReduceOne(int ind)
{
int i, j;
int dim[MAXDIMENS]; //dimensions of the new table
int ndim; //number of dimensions for the new table
int NotFinished = 1;
int* myIndex = new int[nDimens];CheckPointer(myIndex);
memset(myIndex,0,nDimens*sizeof(int));
double s;
LPTable ctab = new Table;
if(NULL == ctab)
{
printf("Error creating new table :: ReduceOne.\n");
return(NULL);
}
ndim = nDimens - 1;
j = 0;
for(i=0; i<nDimens; i++)
{
if(i != ind)
{
dim[j] = Dimens[i];
j++;
}
}
//allocate storage for the new table
ctab->Alloc(dim, ndim);
ctab->GetFirst();
while(NotFinished)
{
j = 0;
for(i=0; i<nDimens; i++)
{
if(i != ind)
{
myIndex[i] = ctab->Index[j];
j++;
}
}
myIndex[ind] = 0;
s = 0.0;
for(i=0; i<Dimens[ind]; i++)
{
s += GetI(myIndex);
myIndex[ind]++;
}
ctab->Set(s);
NotFinished = ctab->GetNext();
}
delete[] myIndex;
return(ctab);
}
void ThetafromP(LPTable theta,LPTable p,int* aModel)
{
int i,j;
double s;
theta->GetFirst();
theta->Set(log(p->Data[0]));
j = 1;
while(theta->GetNext())
{
if(aModel[j])
{
int lenE = 0;
for(i=0;i<theta->nDimens;i++) lenE+=theta->Index[i];
p->GetFirst();
s = pow(-1,lenE)*log(p->Data[0]);
while(p->GetNext())
{
if(subset(p->nDimens,p->Index,theta->Index))
{
int lenF = 0;
for(i=0;i<p->nDimens;i++) lenF+=p->Index[i];
s+= pow(-1.0,lenE-lenF)*log(p->Get());
}
}
}
else
{
s = 0;
}
theta->Set(s);
j++;
}
return;
}
void FindHierarchicalModels(CRegression* allregs,CData& Data,char* DataFileName,int target,int nMaxRegressors)
{
int i,j;
char buffer[2048];
FILE* out = NULL;
int responseVariable = target-1;
int lenmodel = 1+nMaxRegressors + Data.NumOfConfoundingVars; // + all confounding vars
int* amodel = new int[lenmodel];
CRegression* p = allregs->Next;
int modelid = 1;
while(NULL!=p)
{
set<int>::iterator it;
lenmodel = 0;
for(it=(p->Vars).begin();it!=(p->Vars).end();it++)
{
amodel[lenmodel] = *it;
lenmodel++;
}
// Next insert all the confounding vars, followed by the response var
j = Data.NumOfConfoundingVars;
for(i=0; i<=j; i++)
{
amodel[lenmodel] = responseVariable-j+i;
lenmodel++;
}
//amodel[lenmodel] = responseVariable;
//lenmodel++;
printf("Model [%d] ::",modelid);
for(i=0;i<lenmodel;i++) printf(" %d",amodel[i]);
printf("\n");
//get the marginal table for these variables
LPTable dataTable = new Table;
int* index = new int[lenmodel];
for(j=0;j<lenmodel;j++)
{
index[j] = 2;
}
if(!dataTable->Alloc(index,lenmodel))
{
printf("Error allocating memory!\n");
exit(1);
}
for(i=0;i<Data.SampleSize;i++)
{
for(j=0;j<lenmodel;j++)
{
index[j] = (int) (Data.data[i][amodel[j]]);
}
dataTable->SetIndex(index);
dataTable->Set(dataTable->Get()+1);
}
//file where the best model will be saved
sprintf(buffer,"%s.shotgun.%d.%d.reg.model%d.txt",DataFileName,target,nMaxRegressors,modelid);
if(NULL==(out=fopen(buffer,"w")))
{
printf("Cannot open file [%s]\n",buffer);
return;
}
//do the shotgun search
TableShotgunSearch(out, dataTable, model.mnShotgunChainReplicates,
model.mdShotgunCutoffMax, model.mdShotgunCutoffMin, model.mdShotgunProbMax, Data.NumOfConfoundingVars);
fclose(out);
//clean memory
dataTable->Reset();
delete dataTable; dataTable = NULL;
//go to the next model
p = p->Next;
modelid++;
}
//clean memory
delete[] amodel; amodel = NULL;
return;
}
void getNextTheta(int** VarSets,int* lenVarSets,int nVarSets,
int* amodel,int lenCL,int* CL,LPNTable smalltheta,LPTable Theta,LPTable nextTheta)
{
int i,iF;
int okay1;
int iE, iL, iC;
const int NTD = nextTheta->nDimens;
int* CLcomplement = new int[NTD];
int* CLfull = new int[NTD];
int* FunionL = new int[NTD];
const int TT = Theta->Total;
double g[TT];
int len[TT];
for(i=0;i<NTD;i++)
{
CLcomplement[i] = 1;
CLfull[i] = 0;
}
for(i=0;i<lenCL;i++)
{
CLcomplement[CL[i]] = 0;
CLfull[CL[i]] = 1;
}
for(i=0;i<nextTheta->Total;i++)
{
nextTheta->Data[i] = Theta->Data[i];
g[i] = 0;
len[i]=0;
}
////////////////////////////////////////////////
Theta->GetFirst();
okay1 = 1;
iF = 0;
while(okay1)
{
len[iF] = lenVarSets[iF];
double s1 = 1.0;
for(iL=1;iL<nVarSets;iL++)
{
if(subset(Theta->nDimens,VarSets[iL],CLcomplement))
{
for(i=0;i<Theta->nDimens;i++)
{
FunionL[i] = 0;
if(VarSets[iL][i]==1) FunionL[i] = 1;
if(Theta->Index[i]==1) FunionL[i] = 1;
}
int thereisone = 0;
double sC = 0.0;
for(iC=1;iC<nVarSets;iC++)
{
if(0==subset(Theta->nDimens,VarSets[iC],Theta->Index))
{
if(1==subset(Theta->nDimens,VarSets[iC],FunionL))
{
sC += Theta->GetI(VarSets[iC]);
thereisone = 1;
}
}
}
if(thereisone)
{
s1 += exp(sC);
}
}
}
g[iF]=log(s1);
iF++;
okay1 = Theta->GetNext();
}
////////////////////////////////////////////////
nextTheta->GetFirst();
nextTheta->Set(0);
iE = 0;
while(nextTheta->GetNext())
{
iE++;
if(amodel[iE])
{
if(1==subset(Theta->nDimens,nextTheta->Index,CLfull))
{
double s0 = smalltheta->GetI(nextTheta->Index);
int lenE = 0;
for(i=0;i<NTD;i++) lenE+=nextTheta->Index[i];
Theta->GetFirst();
int okay1 = 1;
iF = 0;
while(okay1)
{
if(len[iF]>=0)
{
if(subset(Theta->nDimens,Theta->Index,nextTheta->Index))
{
s0 += pow(-1,lenE-len[iF]-1)*g[iF];
}
}
iF++;
okay1 = Theta->GetNext();
}
nextTheta->Set(s0);
}
else
{
Theta->SetIndex(nextTheta->Index);
nextTheta->Set(Theta->Get());
}
}
else
{
nextTheta->Set(0);
}
}
delete[] CLcomplement; CLcomplement = NULL;
delete[] CLfull; CLfull = NULL;
delete[] FunionL; FunionL = NULL;
return;
}
LPTable ReduceOneShuttle(LPTable tab, int ind, LPTable tabS)
{
LPTable tabR = tab->ReduceOne(ind);
if(NULL == tabR)
{
printf("Error creating new table :: ReduceOneShuttle.\n");
return(NULL);
}
LPTable rez = new Table;
if(NULL == rez)
{
printf("Error creating new table :: ReduceOneShuttle.\n");
tabR->Reset(); delete tabR;
return(NULL);
}
if(!rez->Alloc(tab->Dimens, tab->nDimens))
{
printf("Error creating new table :: ReduceOneShuttle.\n");
tabR->Reset(); delete tabR;
delete rez;
return(NULL);
}
rez->GetFirst();
int i, j;
int NotFinished = 1;
while(NotFinished)
{
j = 0;
for(i=0; i<rez->nDimens; i++)
{
if(i != ind)
{
tabR->Index[j] = rez->Index[i];
j++;
tabS->Index[i] = rez->Index[i];
}
}
double s = 0.0;
for(i=0; i<rez->Dimens[ind]; i++)
{
if(i != rez->Index[ind])
{
tabS->Index[ind] = i;
s += tabS->Get();
}
}
rez->Set(tabR->Get() - s);
if(NotFinished)
{
NotFinished = rez->GetNext();
}
}
tabR->Reset(); delete tabR;
return rez;
}