Example #1
0
qFinderDMM::qFinderDMM(vector<vector<int> > cm, int p): countMatrix(cm), numPartitions(p){
    try {
        m = MothurOut::getInstance();
        numSamples = (int)countMatrix.size();
        numOTUs = (int)countMatrix[0].size();
        
        
        kMeans();
        //    cout << "done kMeans" << endl;
        
        optimizeLambda();
        
        
        //    cout << "done optimizeLambda" << endl;
        
        double change = 1.0000;
        currNLL = 0.0000;
        
        int iter = 0;
        
        while(change > 1.0e-6 && iter < 100){
            
            //        cout << "Calc_Z: ";
            calculatePiK();
            
            optimizeLambda();
            
            //        printf("Iter:%d\t",iter);
            
            for(int i=0;i<numPartitions;i++){
                weights[i] = 0.0000;
                for(int j=0;j<numSamples;j++){
                    weights[i] += zMatrix[i][j];
                }
                //            printf("w_%d=%.3f\t",i,weights[i]);
                
            }
            
            double nLL = getNegativeLogLikelihood();
            
            change = abs(nLL - currNLL);
            
            currNLL = nLL;
            
            //        printf("NLL=%.5f\tDelta=%.4e\n",currNLL, change);
            
            iter++;
        }
        
        error.resize(numPartitions);
        
        logDeterminant = 0.0000;
        
        LinearAlgebra l;
        
        for(currentPartition=0;currentPartition<numPartitions;currentPartition++){
            
            error[currentPartition].assign(numOTUs, 0.0000);
            
            if(currentPartition > 0){
                logDeterminant += (2.0 * log(numSamples) - log(weights[currentPartition]));
            }
            vector<vector<double> > hessian = getHessian();
            vector<vector<double> > invHessian = l.getInverse(hessian);
            
            for(int i=0;i<numOTUs;i++){
                logDeterminant += log(abs(hessian[i][i]));
                error[currentPartition][i] = invHessian[i][i];
            }
            
        }
        
        int numParameters = numPartitions * numOTUs + numPartitions - 1;
        laplace = currNLL + 0.5 * logDeterminant - 0.5 * numParameters * log(2.0 * 3.14159);
        bic = currNLL + 0.5 * log(numSamples) * numParameters;
        aic = currNLL + numParameters;
    }
	catch(exception& e) {
		m->errorOut(e, "qFinderDMM", "qFinderDMM");
		exit(1);
	}
}
Example #2
0
qFinderDMM::qFinderDMM(vector<vector<int> > cm, int p) : CommunityTypeFinder() {
    try {
        //cout << "here" << endl;
        numPartitions = p;
        countMatrix = cm;
        numSamples = (int)countMatrix.size();
        numOTUs = (int)countMatrix[0].size();
        
       // if (m->debug) { m->mothurOut("before kmeans\n"); }
        findkMeans();
       //if (m->debug) { m->mothurOut("done kMeans\n"); }
        
        optimizeLambda();
        
        
         //if (m->debug) { m->mothurOut("done optimizeLambda\n"); }
        
        double change = 1.0000;
        currNLL = 0.0000;
        
        int iter = 0;
        
        while(change > 1.0e-6 && iter < 100){
            
           // if (m->debug) { m->mothurOut("Calc_Z: \n"); }
            calculatePiK();
            
            optimizeLambda();
            
              // if (m->debug) { m->mothurOut("Iter: " + toString(iter) + "\n"); }
            
            for(int i=0;i<numPartitions;i++){
                weights[i] = 0.0000;
                for(int j=0;j<numSamples;j++){
                    weights[i] += zMatrix[i][j];
                }
                           // printf("w_%d=%.3f\t",i,weights[i]);
                
            }
            
            double nLL = getNegativeLogLikelihood();
            
            change = abs(nLL - currNLL);
            
            currNLL = nLL;
            
                  //  printf("NLL=%.5f\tDelta=%.4e\n",currNLL, change);
            
            iter++;
        }
        //if (m->debug) { m->mothurOut("done while loop\n"); }
        error.resize(numPartitions);
        
        logDeterminant = 0.0000;
        
        LinearAlgebra l;
        
        for(currentPartition=0;currentPartition<numPartitions;currentPartition++){
            
            error[currentPartition].assign(numOTUs, 0.0000);
            
            if (m->debug) { m->mothurOut("current partition = " + toString(currentPartition) + "\n"); }
            
            if(currentPartition > 0){
                logDeterminant += (2.0 * log(numSamples) - log(weights[currentPartition]));
            }
            //if (m->debug) { m->mothurOut("before hession\n"); }
            vector<vector<double> > hessian = getHessian();
            //if (m->debug) { m->mothurOut("after hession\n"); }
            vector<vector<double> > invHessian = l.getInverse(hessian);
            //if (m->debug) { m->mothurOut("after inverse\n"); }
            for(int i=0;i<numOTUs;i++){
                logDeterminant += log(abs(hessian[i][i]));
                error[currentPartition][i] = invHessian[i][i];
            }
        }
        
        int numParameters = numPartitions * numOTUs + numPartitions - 1;
        laplace = currNLL + 0.5 * logDeterminant - 0.5 * numParameters * log(2.0 * 3.14159);
        bic = currNLL + 0.5 * log(numSamples) * numParameters;
        aic = currNLL + numParameters;
    }
	catch(exception& e) {
		m->errorOut(e, "qFinderDMM", "qFinderDMM");
		exit(1);
	}
}