public double merge(
int index1,
int index2,
int clusterIndex1,
int clusterIndex2,
int[] cluster1Sites,
int[] cluster2Sites) {
double logqMerge = 0.0;
HashMap<Integer, Integer> siteMap = new HashMap<Integer, Integer>();
// The value of the merged cluster will have that of cluster 2 before the merge.
QuietRealParameter mergedParam = paramList.getParameter(clusterIndex2);
QuietRealParameter mergedModel = modelList.getParameter(clusterIndex2);
QuietRealParameter mergedFreqs = freqsList.getParameter(clusterIndex2);
QuietRealParameter mergedRates = ratesList.getParameter(clusterIndex2);
// Create a vector that combines the site indices of the two clusters
int[] mergedClusterSites = new int[cluster1Sites.length + cluster2Sites.length - 2];
int k = 0;
for (int i = 0; i < cluster1Sites.length; i++) {
// Point every member in cluster 1 to cluster 2
// paramPointers.point(cluster1Sites[i],mergedParam);
// modelPointers.point(cluster1Sites[i],mergedModel);
// freqsPointers.point(cluster1Sites[i],mergedFreqs);
// ratesPointers.point(cluster1Sites[i],mergedRates);
if (cluster1Sites[i] != index1) {
// For all members that are not index 1,
// record the cluster in which they have been before the merge,
// and assign them to the combined vector.
siteMap.put(cluster1Sites[i], clusterIndex1);
mergedClusterSites[k++] = cluster1Sites[i];
}
}
for (int i = 0; i < cluster2Sites.length; i++) {
// All members in cluster 2 remains in cluster2 so no new pointer assignments
if (cluster2Sites[i] != index2) {
// For all members that are not index 2,
// record the cluster in which they have been before the merge,
// and assign them to the combined vector.
siteMap.put(cluster2Sites[i], clusterIndex2);
mergedClusterSites[k++] = cluster2Sites[i];
}
}
try {
// Create a weight vector of patterns to inform the temporary tree likelihood
// which set of pattern likelihoods are to be computed.
// int[] tempWeights = dpTreeLikelihood.getClusterWeights(clusterIndex1);
/*int[] tempWeights = new int[tempLikelihood.m_data.get().getPatternCount()];
for(int i = 0; i < cluster1Sites.length; i++){
int patIndex = tempLikelihood.m_data.get().getPatternIndex(cluster1Sites[i]);
tempWeights[patIndex] = 1;
}
tempLikelihood.setPatternWeights(tempWeights);
double[] cluster1SitesCluster2ParamLogLik = tempLikelihood.calculateLogP(
mergedParam,
mergedModel,
mergedFreqs,
mergedRates,
cluster1Sites,
index1
); */
k = 0;
int[] sCluster1Sites = new int[cluster1Sites.length - 1];
for (int i = 0; i < cluster1Sites.length; i++) {
if (cluster1Sites[i] != index1) {
sCluster1Sites[k++] = cluster1Sites[i];
}
}
tempLikelihood.setupPatternWeightsFromSites(sCluster1Sites);
double[] cluster1SitesCluster2ParamLogLik =
tempLikelihood.calculateLogP(
mergedParam, mergedModel, mergedFreqs, mergedRates, sCluster1Sites);
// tempWeights = dpTreeLikelihood.getClusterWeights(clusterIndex2);
/*tempWeights = new int[tempLikelihood.m_data.get().getPatternCount()];
for(int i = 0; i < cluster2Sites.length; i++){
int patIndex = tempLikelihood.m_data.get().getPatternIndex(cluster2Sites[i]);
tempWeights[patIndex] = 1;
}
tempLikelihood.setPatternWeights(tempWeights);
RealParameter removedParam = paramList.getParameter(clusterIndex1);
RealParameter removedModel = modelList.getParameter(clusterIndex1);
RealParameter removedFreqs = freqsList.getParameter(clusterIndex1);
RealParameter removedRates = ratesList.getParameter(clusterIndex1);
double[] cluster2SitesCluster1ParamLogLik = tempLikelihood.calculateLogP(
removedParam,
removedModel,
removedFreqs,
removedRates,
cluster2Sites,
index2
);*/
k = 0;
int[] sCluster2Sites = new int[cluster2Sites.length - 1];
for (int i = 0; i < cluster2Sites.length; i++) {
if (cluster2Sites[i] != index2) {
sCluster2Sites[k++] = cluster2Sites[i];
}
}
tempLikelihood.setupPatternWeightsFromSites(sCluster2Sites);
RealParameter removedParam = paramList.getParameter(clusterIndex1);
RealParameter removedModel = modelList.getParameter(clusterIndex1);
RealParameter removedFreqs = freqsList.getParameter(clusterIndex1);
RealParameter removedRates = ratesList.getParameter(clusterIndex1);
double[] cluster2SitesCluster1ParamLogLik =
tempLikelihood.calculateLogP(
removedParam, removedModel, removedFreqs, removedRates, sCluster2Sites);
// System.out.println("populate logLik1:");
double[] logLik1 = new double[mergedClusterSites.length];
for (int i = 0; i < (cluster1Sites.length - 1); i++) {
// System.out.println(clusterIndex1+" "+mergedClusterSites[i]);
logLik1[i] = dpTreeLikelihood.getSiteLogLikelihood(clusterIndex1, mergedClusterSites[i]);
}
System.arraycopy(
cluster2SitesCluster1ParamLogLik,
0,
logLik1,
cluster1Sites.length - 1,
cluster2SitesCluster1ParamLogLik.length);
double[] logLik2 = new double[mergedClusterSites.length];
System.arraycopy(
cluster1SitesCluster2ParamLogLik, 0, logLik2, 0, cluster1SitesCluster2ParamLogLik.length);
// System.out.println("populate logLik2:");
for (int i = cluster1SitesCluster2ParamLogLik.length; i < logLik2.length; i++) {
// System.out.println(clusterIndex2+"
// "+mergedClusterSites[i-cluster1SitesCluster2ParamLogLik.length]);
logLik2[i] = dpTreeLikelihood.getSiteLogLikelihood(clusterIndex2, mergedClusterSites[i]);
}
double[] lik1 = new double[logLik1.length];
double[] lik2 = new double[logLik2.length];
// scale it so it may be more accuate
double minLog;
/*for(int i = 0; i < logLik1.length; i++){
minLog = Math.min(logLik1[i],logLik2[i]);
if(minLog == logLik1[i]){
lik1[i] = 1.0;
lik2[i] = Math.exp(logLik2[i] - minLog);
}else{
lik1[i] = Math.exp(logLik1[i] - minLog);
lik2[i] = 1.0;
}
}*/
for (int i = 0; i < logLik1.length; i++) {
lik1[i] = Math.exp(logLik1[i]);
lik2[i] = Math.exp(logLik2[i]);
// System.out.println(lik1[i]+" "+lik2[i]);
}
// Create a set of indices for random permutation
int[] shuffle = new int[mergedClusterSites.length];
for (int i = 0; i < shuffle.length; i++) {
shuffle[i] = i;
}
Randomizer.shuffle(shuffle);
int cluster1Count = 1;
int cluster2Count = 1;
int cluster;
double psi1, psi2, cluster1Prob;
for (int i = 0; i < mergedClusterSites.length; i++) {
cluster = siteMap.get(mergedClusterSites[shuffle[i]]);
psi1 = cluster1Count * lik1[shuffle[i]];
psi2 = cluster2Count * lik2[shuffle[i]];
/*testCorrectness(i,cluster,
clusterIndex1,clusterIndex2,shuffle, mergedClusterSites,
lik1,lik2);*/
cluster1Prob = psi1 / (psi1 + psi2);
// System.out.println(cluster1Prob);
if (cluster == clusterIndex1) {
logqMerge += Math.log(cluster1Prob);
cluster1Count++;
} else if (cluster == clusterIndex2) {
logqMerge += Math.log(1 - cluster1Prob);
cluster2Count++;
} else {
throw new RuntimeException("Something is wrong.");
}
}
logqMerge += // paramBaseDistr.calcLogP(removedParam)
mergeValue(removedParam, mergedParam, paramBaseDistr)
// + modelBaseDistr.calcLogP(removedModel)
+ mergeDiscreteValue(removedModel, mergedModel, modelDistrInput.get())
+ freqsBaseDistr.calcLogP(removedFreqs)
// + ratesBaseDistr.calcLogP(removedRates);
+ mergeValueInLogSpace(removedRates, mergedRates, ratesBaseDistr);
if (logqMerge > Double.NEGATIVE_INFINITY) {
paramList.mergeParameter(clusterIndex1, clusterIndex2);
modelList.mergeParameter(clusterIndex1, clusterIndex2);
freqsList.mergeParameter(clusterIndex1, clusterIndex2);
ratesList.mergeParameter(clusterIndex1, clusterIndex2);
for (int i = 0; i < cluster1Sites.length; i++) {
// Point every member in cluster 1 to cluster 2
paramPointers.point(cluster1Sites[i], mergedParam);
modelPointers.point(cluster1Sites[i], mergedModel);
freqsPointers.point(cluster1Sites[i], mergedFreqs);
ratesPointers.point(cluster1Sites[i], mergedRates);
}
}
} catch (Exception e) {
throw new RuntimeException(e);
}
return logqMerge;
}
