/** For each point p, calculate function Kdist(p) which is the distance of
* the Kth nearest point to p.
*/
void Cluster_DBSCAN::ComputeKdist( int Kval, std::vector<int> const& FramesToCluster ) const {
std::vector<double> dists;
std::vector<double> Kdist;
dists.reserve( FramesToCluster.size() );
Kdist.reserve( FramesToCluster.size() );
std::string outfilename = k_prefix_ + "Kdist." + integerToString(Kval) + ".dat";
mprintf("\tDBSCAN: Calculating Kdist(%i), output to %s\n", Kval, outfilename.c_str());
for (std::vector<int>::const_iterator point = FramesToCluster.begin();
point != FramesToCluster.end();
++point)
{
// Store distances from this point
dists.clear();
for (std::vector<int>::const_iterator otherpoint = FramesToCluster.begin();
otherpoint != FramesToCluster.end();
++otherpoint)
dists.push_back( FrameDistances_.GetFdist(*point, *otherpoint) );
// Sort distances - first dist should always be 0
std::sort(dists.begin(), dists.end());
Kdist.push_back( dists[Kval] );
}
std::sort( Kdist.begin(), Kdist.end() );
CpptrajFile Outfile;
Outfile.OpenWrite(outfilename);
Outfile.Printf("%-8s %1i%-11s\n", "#Point", Kval,"-dist");
// Write out largest to smallest
unsigned int ik = 0;
for (std::vector<double>::reverse_iterator k = Kdist.rbegin();
k != Kdist.rend(); ++k, ++ik)
Outfile.Printf("%8u %12.4f\n", ik, *k);
Outfile.CloseFile();
}
// DataIO_Std::WriteData()
int DataIO_Std::WriteData(FileName const& fname, DataSetList const& SetList)
{
int err = 0;
if (!SetList.empty()) {
// Open output file.
CpptrajFile file;
if (file.OpenWrite( fname )) return 1;
// Base write type off first data set dimension FIXME
if (SetList[0]->Group() == DataSet::CLUSTERMATRIX) {
// Special case of 2D - may have sieved frames.
err = WriteCmatrix(file, SetList);
} else if (SetList[0]->Ndim() == 1) {
if (group_ == NO_TYPE) {
if (isInverted_)
err = WriteDataInverted(file, SetList);
else
err = WriteDataNormal(file, SetList);
} else
err = WriteByGroup(file, SetList, group_);
} else if (SetList[0]->Ndim() == 2)
err = WriteData2D(file, SetList);
else if (SetList[0]->Ndim() == 3)
err = WriteData3D(file, SetList);
file.CloseFile();
}
return err;
}
// TODO: Accept const ArgList so arguments are not reset?
CpptrajFile* DataFileList::AddCpptrajFile(FileName const& nameIn,
std::string const& descrip,
CFtype typeIn, bool allowStdout)
{
// If no filename and stdout not allowed, no output desired.
if (nameIn.empty() && !allowStdout) return 0;
FileName name;
CpptrajFile* Current = 0;
int currentIdx = -1;
if (!nameIn.empty()) {
name = nameIn;
// Append ensemble number if set.
if (ensembleNum_ != -1)
name.Append( "." + integerToString(ensembleNum_) );
// Check if filename in use by DataFile.
DataFile* df = GetDataFile(name);
if (df != 0) {
mprinterr("Error: Text output file name '%s' already in use by data file '%s'.\n",
nameIn.full(), df->DataFilename().full());
return 0;
}
// Check if this filename already in use
currentIdx = GetCpptrajFileIdx( name );
if (currentIdx != -1) Current = cfList_[currentIdx];
}
// If no CpptrajFile associated with name, create new CpptrajFile
if (Current==0) {
switch (typeIn) {
case TEXT: Current = new CpptrajFile(); break;
case PDB: Current = (CpptrajFile*)(new PDBfile()); break;
}
Current->SetDebug(debug_);
// Set up file for writing.
//if (Current->SetupWrite( name, debug_ ))
if (Current->OpenWrite( name ))
{
mprinterr("Error: Setting up text output file %s\n", name.full());
delete Current;
return 0;
}
cfList_.push_back( Current );
cfData_.push_back( CFstruct(descrip, typeIn) );
} else {
// If Current type does not match typeIn do not allow.
if (typeIn != cfData_[currentIdx].Type()) {
mprinterr("Error: Cannot change type of text output for '%s'.\n", Current->Filename().full());
return 0;
}
Current->SetDebug(debug_);
// Update description
if (!descrip.empty())
cfData_[currentIdx].UpdateDescrip( descrip );
}
return Current;
}
// DataIO_CCP4::WriteData()
int DataIO_CCP4::WriteData(FileName const& fname, DataSetList const& setList)
{
// Open output file
CpptrajFile outfile;
if (outfile.OpenWrite(fname)) {
mprinterr("Error: Could not open CCP4 output file '%s'.\n", fname.full());
return 1;
}
// Warn about writing multiple sets
if (setList.size() > 1)
mprintf("Warning: %s: Writing multiple 3D sets in CCP4 format not supported.\n"
"Warning: Only writing first set.\n", fname.full());
return WriteSet3D( setList.begin(), outfile );
}
// DataIO_OpenDx::WriteData()
int DataIO_OpenDx::WriteData(FileName const& fname, DataSetList const& setList)
{
// Open output file
CpptrajFile outfile;
if (outfile.OpenWrite(fname)) {
mprinterr("Error: Could not open OpenDX output file.\n");
return 1;
}
// Warn about writing multiple sets
if (setList.size() > 1)
mprintf("Warning: %s: Writing multiple 3D sets in OpenDX format may result in unexpected behavior\n", fname.full());
int err = 0;
for (DataSetList::const_iterator set = setList.begin(); set != setList.end(); ++set)
err += WriteSet3D( *(*set), outfile );
return err;
}
void Action_Pairwise::Print() {
if (nframes_ < 1) return;
// Divide matrices by # of frames
double norm = 1.0 / (double)nframes_;
for (unsigned int i = 0; i != vdwMat_->Size(); i++)
{
(*vdwMat_)[i] *= norm;
(*eleMat_)[i] *= norm;
}
// Write out final results
CpptrajFile AvgOut;
if (AvgOut.OpenWrite( avgout_ )) return;
if (nb_calcType_ == NORMAL)
mprintf(" PAIRWISE: Writing all pairs with |<evdw>| > %.4f, |<eelec>| > %.4f\n",
cut_evdw_, cut_eelec_);
else if (nb_calcType_ == COMPARE_REF)
mprintf(" PAIRWISE: Writing all pairs with |<dEvdw>| > %.4f, |<dEelec>| > %.4f\n",
cut_evdw_, cut_eelec_);
AvgOut.Printf("%-16s %5s -- %16s %5s : ENE\n","#Name1", "At1", "Name2", "At2");
for (AtomMask::const_iterator m1 = Mask0_.begin(); m1 != Mask0_.end(); ++m1) {
for (AtomMask::const_iterator m2 = m1 + 1; m2 != Mask0_.end(); ++m2)
{
double EV = vdwMat_->GetElement(*m1, *m2);
double EE = eleMat_->GetElement(*m1, *m2);
bool outputv = ( fabs(EV) > cut_evdw_ );
bool outpute = ( fabs(EE) > cut_eelec_ );
if (outputv || outpute) {
AvgOut.Printf("%16s %5i -- %16s %5i :",
CurrentParm_->TruncResAtomName(*m1).c_str(), *m1 + 1,
CurrentParm_->TruncResAtomName(*m2).c_str(), *m2 + 1);
if (outputv) AvgOut.Printf(" EVDW= %12.5e", EV);
if (outpute) AvgOut.Printf(" EELEC= %12.5e", EE);
AvgOut.Printf("\n");
}
}
}
}
int Cluster_DPeaks::ChoosePointsAutomatically() {
// Right now all density values are discrete. Try to choose outliers at each
// value for which there is density.;
/*
// For each point, calculate average distance (X,Y) to points in next and
// previous density values.
const double dens_cut = 3.0 * 3.0;
const double dist_cut = 1.32 * 1.32;
for (Carray::const_iterator point0 = Points_.begin(); point0 != Points_.end(); ++point0)
{
int Npts = 0;
for (Carray::const_iterator point1 = Points_.begin(); point1 != Points_.end(); ++point1)
{
if (point0 != point1) {
// Only do this for close densities
double dX = (double)(point0->PointsWithinEps() - point1->PointsWithinEps());
double dX2 = dX * dX;
double dY = (point0->Dist() - point1->Dist());
double dY2 = dY * dY;
if (dX2 < dens_cut && dY2 < dist_cut) {
Npts++;
}
}
}
mprintf("%i %i %i\n", point0->PointsWithinEps(), point0->Fnum()+1, Npts);
}
*/
/*
CpptrajFile tempOut;
tempOut.OpenWrite("temp.dat");
int currentDensity = -1;
double distAv = 0.0;
double distSD = 0.0;
double sumWts = 0.0;
int nValues = 0;
Carray::const_iterator lastPoint = Points_.end() + 1;
for (Carray::const_iterator point = Points_.begin(); point != lastPoint; ++point)
{
if (point == Points_.end() || point->PointsWithinEps() != currentDensity) {
if (nValues > 0) {
distAv = distAv / sumWts; //(double)nValues;
distSD = (distSD / sumWts) - (distAv * distAv);
if (distSD > 0.0)
distSD = sqrt(distSD);
else
distSD = 0.0;
//mprintf("Density %i: %i values Avg= %g SD= %g SumWts= %g\n", currentDensity,
// nValues, distAv, distSD, sumWts);
tempOut.Printf("%i %g\n", currentDensity, distAv);
}
if (point == Points_.end()) break;
currentDensity = point->PointsWithinEps();
distAv = 0.0;
distSD = 0.0;
sumWts = 0.0;
nValues = 0;
}
double wt = exp(point->Dist());
double dval = point->Dist() * wt;
sumWts += wt;
distAv += dval;
distSD += (dval * dval);
nValues++;
}
tempOut.CloseFile();
*/
// BEGIN CALCULATING WEIGHTED DISTANCE AVERAGE
CpptrajFile tempOut;
tempOut.OpenWrite("temp.dat");
DataSet_Mesh weightedAverage;
Carray::const_iterator cp = Points_.begin();
// Skip local density of 0.
//while (cp->PointsWithinEps() == 0 && cp != Points_.end()) ++cp;
while (cp != Points_.end())
{
int densityVal = cp->PointsWithinEps();
Carray densityArray;
// Add all points of current density.
while (cp->PointsWithinEps() == densityVal && cp != Points_.end())
densityArray.push_back( *(cp++) );
mprintf("Density value %i has %zu points.\n", densityVal, densityArray.size());
// Sort array by distance
std::sort(densityArray.begin(), densityArray.end(), Cpoint::dist_sort());
// Take the average of the points weighted by their position.
double wtDistAv = 0.0;
double sumWts = 0.0;
//std::vector<double> weights;
//weights.reserve( densityArray.size() );
int maxPt = (int)densityArray.size() - 1;
for (int ip = 0; ip != (int)densityArray.size(); ++ip)
{
double wt = exp( (double)(ip - maxPt) );
//mprintf("\t%10i %10u %10u %10g\n", densityVal, ip, maxPt, wt);
wtDistAv += (densityArray[ip].Dist() * wt);
sumWts += wt;
//weights.push_back( wt );
}
wtDistAv /= sumWts;
// Calculate the weighted sample variance
//double distSD = 0.0;
//for (unsigned int ip = 0; ip != densityArray.size(); ++ip) {
// double diff = densityArray[ip].Dist() - wtDistAv;
// distSD += weights[ip] * (diff * diff);
//}
//distSD /= sumWts;
weightedAverage.AddXY(densityVal, wtDistAv);
//tempOut.Printf("%i %g %g %g\n", densityVal, wtDistAv, sqrt(distSD), sumWts);
tempOut.Printf("%i %g %g\n", densityVal, wtDistAv, sumWts);
/*
// Find the median.
double median, Q1, Q3;
if (densityArray.size() == 1) {
median = densityArray[0].Dist();
Q1 = median;
Q3 = median;
} else {
unsigned int q3_beg;
unsigned int med_idx = densityArray.size() / 2; // Always 0 <= Q1 < med_idx
if ((densityArray.size() % 2) == 0) {
median = (densityArray[med_idx].Dist() + densityArray[med_idx-1].Dist()) / 2.0;
q3_beg = med_idx;
} else {
median = densityArray[med_idx].Dist();
q3_beg = med_idx + 1;
}
if (densityArray.size() == 2) {
Q1 = densityArray[0].Dist();
Q3 = densityArray[1].Dist();
} else {
// Find lower quartile
unsigned int q1_idx = med_idx / 2;
if ((med_idx % 2) == 0)
Q1 = (densityArray[q1_idx].Dist() + densityArray[q1_idx-1].Dist()) / 2.0;
else
Q1 = densityArray[q1_idx].Dist();
// Find upper quartile
unsigned int q3_size = densityArray.size() - q3_beg;
unsigned int q3_idx = (q3_size / 2) + q3_beg;
if ((q3_size %2) == 0)
Q3 = (densityArray[q3_idx].Dist() + densityArray[q3_idx-1].Dist()) / 2.0;
else
Q3 = densityArray[q3_idx].Dist();
}
}
mprintf("\tMedian dist value is %g. Q1= %g Q3= %g\n", median, Q1, Q3);
*/
}
tempOut.CloseFile();
// END CALCULATING WEIGHTED DISTANCE AVERAGE
/*
// TEST
tempOut.OpenWrite("temp2.dat");
std::vector<double> Hist( Points_.back().PointsWithinEps()+1, 0.0 );
int gWidth = 3;
double cval = 3.0;
double two_c_squared = 2.0 * cval * cval;
mprintf("DBG: cval= %g, Gaussian denominator is %g\n", cval, two_c_squared);
for (int wtIdx = 0; wtIdx != (int)weightedAverage.Size(); wtIdx++)
{
int bval = weightedAverage.X(wtIdx);
for (int xval = std::max(bval - gWidth, 0);
xval != std::min(bval + gWidth + 1, (int)Hist.size()); xval++)
{
// a: height (weighted average)
// b: center (density value)
// c: width
// x: density value in histogram
//int xval = weightedAverage.X(idx);
//double bval = weightedAverage.X(wtIdx);
//double bval = (double)wtIdx;
double diff = (double)(xval - bval);
//Hist[xval] += (weightedAverage.Y(wtIdx) * exp( -( (diff * diff) / two_c_squared ) ));
Hist[xval] = std::max(Hist[xval],
weightedAverage.Y(wtIdx) * exp( -( (diff * diff) / two_c_squared ) ));
}
}
for (unsigned int idx = 0; idx != Hist.size(); idx++)
tempOut.Printf("%u %g\n", idx, Hist[idx]);
tempOut.CloseFile();
// END TEST
*/
/*
// TEST
// Construct best-fit line segments
tempOut.OpenWrite("temp2.dat");
double slope, intercept, correl;
int segment_length = 3;
DataSet_Mesh Segment;
Segment.Allocate1D( segment_length );
for (int wtIdx = 0; wtIdx != (int)weightedAverage.Size(); wtIdx++)
{
Segment.Clear();
for (int idx = std::max(wtIdx - 1, 0); // TODO: use segment_length
idx != std::min(wtIdx + 2, (int)weightedAverage.Size()); idx++)
Segment.AddXY(weightedAverage.X(idx), weightedAverage.Y(idx));
Segment.LinearRegression(slope, intercept, correl, true);
for (int idx = std::max(wtIdx - 1, 0); // TODO: use segment_length
idx != std::min(wtIdx + 2, (int)weightedAverage.Size()); idx++)
{
double x = weightedAverage.X(idx);
double y = slope * x + intercept;
tempOut.Printf("%g %g %i\n", x, y, weightedAverage.X(wtIdx));
}
}
tempOut.CloseFile();
// END TEST
*/
// BEGIN WEIGHTED RUNNING AVG/SD OF DISTANCES
// For each point, determine if it is greater than the average of the
// weighted average distances of the previous, current, and next densities.
int width = 2;
int currentDensity = 0;
int wtIdx = 0;
double currentAvg = 0.0;
double deltaSD = 0.0;
double deltaAv = 0.0;
int Ndelta = 0;
CpptrajFile raOut;
if (!rafile_.empty()) raOut.OpenWrite(rafile_);
CpptrajFile raDelta;
if (!radelta_.empty()) raDelta.OpenWrite(radelta_);
std::vector<unsigned int> candidateIdxs;
std::vector<double> candidateDeltas;
cp = Points_.begin();
// Skip over points with zero density
while (cp != Points_.end() && cp->PointsWithinEps() == 0) ++cp;
while (weightedAverage.X(wtIdx) != cp->PointsWithinEps() && wtIdx < (int)Points_.size())
++wtIdx;
for (Carray::const_iterator point = cp; point != Points_.end(); ++point)
{
if (point->PointsWithinEps() != currentDensity) {
//currentAvg = weightedAverage.Y(wtIdx);
// New density value. Determine average.
currentAvg = 0.0;
// unsigned int Npt = 0;
double currentWt = 0.0;
for (int idx = std::max(wtIdx - width, 0);
idx != std::min(wtIdx + width + 1, (int)weightedAverage.Size()); idx++)
{
//currentAvg += weightedAverage.Y(idx);
//Npt++;
double wt = weightedAverage.Y(idx);
currentAvg += (weightedAverage.Y(idx) * wt);
currentWt += wt;
}
//currentAvg /= (double)Npt;
currentAvg /= currentWt;
//smoothAv += currentAvg;
//smoothSD += (currentAvg * currentAvg);
//Nsmooth++;
currentDensity = point->PointsWithinEps();
if (raOut.IsOpen())
raOut.Printf("%i %g %g\n", currentDensity, currentAvg, weightedAverage.Y(wtIdx));
wtIdx++;
}
double delta = (point->Dist() - currentAvg);
if (delta > 0.0) {
//delta *= log((double)currentDensity);
if (raDelta.IsOpen())
raDelta.Printf("%8i %8.3f %8i %8.3f %8.3f\n",
currentDensity, delta, point->Fnum()+1, point->Dist(), currentAvg);
candidateIdxs.push_back( point - Points_.begin() );
candidateDeltas.push_back( delta );
deltaAv += delta;
deltaSD += (delta * delta);
Ndelta++;
}
}
raOut.CloseFile();
deltaAv /= (double)Ndelta;
deltaSD = (deltaSD / (double)Ndelta) - (deltaAv * deltaAv);
if (deltaSD > 0.0)
deltaSD = sqrt(deltaSD);
else
deltaSD = 0.0;
if (raDelta.IsOpen())
raDelta.Printf("#DeltaAvg= %g DeltaSD= %g\n", deltaAv, deltaSD);
raDelta.CloseFile();
int cnum = 0;
for (unsigned int i = 0; i != candidateIdxs.size(); i++) {
if (candidateDeltas[i] > (deltaSD)) {
Points_[candidateIdxs[i]].SetCluster( cnum++ );
mprintf("\tPoint %u (frame %i, density %i) selected as candidate for cluster %i\n",
candidateIdxs[i], Points_[candidateIdxs[i]].Fnum()+1,
Points_[candidateIdxs[i]].PointsWithinEps(), cnum-1);
}
}
// END WEIGHTED AVG/SD OF DISTANCES
/*
// Currently doing this by calculating the running average of density vs
// distance, then choosing points with distance > twice the SD of the
// running average.
// NOTE: Store in a mesh data set for now in case we want to spline etc later.
if (avg_factor_ < 1) avg_factor_ = 10;
unsigned int window_size = Points_.size() / (unsigned int)avg_factor_;
mprintf("\tRunning avg window size is %u\n", window_size);
// FIXME: Handle case where window_size < frames
DataSet_Mesh runavg;
unsigned int ra_size = Points_.size() - window_size + 1;
runavg.Allocate1D( ra_size );
double dwindow = (double)window_size;
double sumx = 0.0;
double sumy = 0.0;
for (unsigned int i = 0; i < window_size; i++) {
sumx += (double)Points_[i].PointsWithinEps();
sumy += Points_[i].Dist();
}
runavg.AddXY( sumx / dwindow, sumy / dwindow );
for (unsigned int i = 1; i < ra_size; i++) {
unsigned int nextwin = i + window_size - 1;
unsigned int prevwin = i - 1;
sumx = (double)Points_[nextwin].PointsWithinEps() -
(double)Points_[prevwin].PointsWithinEps() + sumx;
sumy = Points_[nextwin].Dist() -
Points_[prevwin].Dist() + sumy;
runavg.AddXY( sumx / dwindow, sumy / dwindow );
}
// Write running average
if (!rafile_.empty()) {
CpptrajFile raOut;
if (raOut.OpenWrite(rafile_))
mprinterr("Error: Could not open running avg file '%s' for write.\n", rafile_.c_str());
else {
for (unsigned int i = 0; i != runavg.Size(); i++)
raOut.Printf("%g %g\n", runavg.X(i), runavg.Y(i));
raOut.CloseFile();
}
}
double ra_sd;
double ra_avg = runavg.Avg( ra_sd );
// Double stdev to use as cutoff for findning anomalously high peaks.
ra_sd *= 2.0;
mprintf("\tAvg of running avg set is %g, SD*2.0 (delta cutoff) is %g\n", ra_avg, ra_sd);
// For each point in density vs distance plot, determine which running
// average point is closest. If the difference between the point and the
// running average point is > 2.0 the SD of the running average data,
// consider it a 'peak'.
CpptrajFile raDelta;
if (!radelta_.empty())
raDelta.OpenWrite("radelta.dat");
if (raDelta.IsOpen())
raDelta.Printf("%-10s %10s %10s\n", "#Frame", "RnAvgPos", "Delta");
unsigned int ra_position = 0; // Position in the runavg DataSet
unsigned int ra_end = runavg.Size() - 1;
int cnum = 0;
for (Carray::iterator point = Points_.begin();
point != Points_.end(); ++point)
{
if (ra_position != ra_end) {
// Is the next running avgd point closer to this point?
while (ra_position != ra_end) {
double dens = (double)point->PointsWithinEps();
double diff0 = fabs( dens - runavg.X(ra_position ) );
double diff1 = fabs( dens - runavg.X(ra_position+1) );
if (diff1 < diff0)
++ra_position; // Next running avg position is closer for this point.
else
break; // This position is closer.
}
}
double delta = point->Dist() - runavg.Y(ra_position);
if (raDelta.IsOpen())
raDelta.Printf("%-10i %10u %10g", point->Fnum()+1, ra_position, delta);
if (delta > ra_sd) {
if (raDelta.IsOpen())
raDelta.Printf(" POTENTIAL CLUSTER %i", cnum);
point->SetCluster(cnum++);
}
if (raDelta.IsOpen()) raDelta.Printf("\n");
}
raDelta.CloseFile();
*/
return cnum;
}
// -----------------------------------------------------------------------------
int Cluster_DPeaks::Cluster_DiscreteDensity() {
mprintf("\tStarting DPeaks clustering, discrete density calculation.\n");
Points_.clear();
// First determine which frames are being clustered.
for (int frame = 0; frame < (int)FrameDistances_.Nframes(); ++frame)
if (!FrameDistances_.IgnoringRow( frame ))
Points_.push_back( Cpoint(frame) );
// Sanity check.
if (Points_.size() < 2) {
mprinterr("Error: Only 1 frame in initial clustering.\n");
return 1;
}
// For each point, determine how many others are within epsilon. Also
// determine maximum distance between any two points.
mprintf("\tDetermining local density of each point.\n");
ProgressBar cluster_progress( Points_.size() );
double maxDist = -1.0;
for (Carray::iterator point0 = Points_.begin();
point0 != Points_.end(); ++point0)
{
cluster_progress.Update(point0 - Points_.begin());
int density = 0;
for (Carray::const_iterator point1 = Points_.begin();
point1 != Points_.end(); ++point1)
{
if (point0 != point1) {
double dist = FrameDistances_.GetFdist(point0->Fnum(), point1->Fnum());
maxDist = std::max(maxDist, dist);
if ( dist < epsilon_ )
density++;
}
}
point0->SetPointsWithinEps( density );
}
mprintf("DBG: Max dist= %g\n", maxDist);
// DEBUG: Frame/Density
CpptrajFile fdout;
fdout.OpenWrite("fd.dat");
for (Carray::const_iterator point = Points_.begin(); point != Points_.end(); ++point)
fdout.Printf("%i %i\n", point->Fnum()+1, point->PointsWithinEps());
fdout.CloseFile();
// Sort by density here. Otherwise array indices will be invalid later.
std::sort( Points_.begin(), Points_.end(), Cpoint::pointsWithinEps_sort() );
// For each point, find the closest point that has higher density. Since
// array is now sorted by density the last point has the highest density.
Points_.back().SetDist( maxDist );
mprintf("\tFinding closest neighbor point with higher density for each point.\n");
unsigned int lastidx = Points_.size() - 1;
cluster_progress.SetupProgress( lastidx );
for (unsigned int idx0 = 0; idx0 != lastidx; idx0++)
{
cluster_progress.Update( idx0 );
double min_dist = maxDist;
int nearestIdx = -1; // Index of nearest neighbor with higher density
Cpoint& point0 = Points_[idx0];
//mprintf("\nDBG:\tSearching for nearest neighbor to idx %u with higher density than %i.\n",
// idx0, point0.PointsWithinEps());
// Since array is sorted by density we can start at the next point.
for (unsigned int idx1 = idx0+1; idx1 != Points_.size(); idx1++)
{
Cpoint const& point1 = Points_[idx1];
double dist1_2 = FrameDistances_.GetFdist(point0.Fnum(), point1.Fnum());
if (point1.PointsWithinEps() > point0.PointsWithinEps())
{
if (dist1_2 < min_dist) {
min_dist = dist1_2;
nearestIdx = (int)idx1;
//mprintf("DBG:\t\tNeighbor idx %i is closer (density %i, distance %g)\n",
// nearestIdx, point1.PointsWithinEps(), min_dist);
}
}
}
point0.SetDist( min_dist );
//mprintf("DBG:\tClosest point to %u with higher density is %i (distance %g)\n",
// idx0, nearestIdx, min_dist);
point0.SetNearestIdx( nearestIdx );
}
// Plot density vs distance for each point.
if (!dvdfile_.empty()) {
CpptrajFile output;
if (output.OpenWrite(dvdfile_))
mprinterr("Error: Could not open density vs distance plot '%s' for write.\n",
dvdfile_.c_str()); // TODO: Make fatal?
else {
output.Printf("%-10s %10s %s %10s %10s\n", "#Density", "Distance",
"Frame", "Idx", "Neighbor");
for (Carray::const_iterator point = Points_.begin();
point != Points_.end(); ++point)
output.Printf("%-10i %10g \"%i\" %10u %10i\n", point->PointsWithinEps(), point->Dist(),
point->Fnum()+1, point-Points_.begin(), point->NearestIdx());
output.CloseFile();
}
}
return 0;
}
// -----------------------------------------------------------------------------
int Cluster_DPeaks::Cluster_GaussianKernel() {
mprintf("\tStarting DPeaks clustering. Using Gaussian kernel to calculate density.\n");
// First determine which frames are being clustered.
Points_.clear();
int oidx = 0;
for (int frame = 0; frame < (int)FrameDistances_.Nframes(); ++frame)
if (!FrameDistances_.IgnoringRow( frame ))
Points_.push_back( Cpoint(frame, oidx++) );
// Sanity check.
if (Points_.size() < 2) {
mprinterr("Error: Only 1 frame in initial clustering.\n");
return 1;
}
// Sort distances
std::vector<float> Distances;
for (ClusterMatrix::const_iterator mat = FrameDistances_.begin();
mat != FrameDistances_.end(); ++mat)
Distances.push_back( *mat );
std::sort( Distances.begin(), Distances.end() );
unsigned int idx = (unsigned int)((double)Distances.size() * 0.02);
double bandwidth = (double)Distances[idx];
mprintf("idx= %u, bandwidth= %g\n", idx, bandwidth);
// Density via Gaussian kernel
double maxDist = -1.0;
for (unsigned int i = 0; i != Points_.size(); i++) {
for (unsigned int j = i+1; j != Points_.size(); j++) {
double dist = FrameDistances_.GetFdist(Points_[i].Fnum(), Points_[j].Fnum());
maxDist = std::max( maxDist, dist );
dist /= bandwidth;
double gk = exp(-(dist *dist));
Points_[i].AddDensity( gk );
Points_[j].AddDensity( gk );
}
}
mprintf("Max dist= %g\n", maxDist);
CpptrajFile rhoOut;
rhoOut.OpenWrite("rho.dat");
for (unsigned int i = 0; i != Points_.size(); i++)
rhoOut.Printf("%u %g\n", i+1, Points_[i].Density());
rhoOut.CloseFile();
// Sort by density, descending
std::stable_sort( Points_.begin(), Points_.end(), Cpoint::density_sort_descend() );
CpptrajFile ordrhoOut;
ordrhoOut.OpenWrite("ordrho.dat");
for (unsigned int i = 0; i != Points_.size(); i++)
ordrhoOut.Printf("%u %g %i %i\n", i+1, Points_[i].Density(), Points_[i].Fnum()+1,
Points_[i].Oidx()+1);
ordrhoOut.CloseFile();
// Determine minimum distances
int first_idx = Points_[0].Oidx();
Points_[first_idx].SetDist( -1.0 );
Points_[first_idx].SetNearestIdx(-1);
for (unsigned int ii = 1; ii != Points_.size(); ii++) {
int ord_i = Points_[ii].Oidx();
Points_[ord_i].SetDist( maxDist );
for (unsigned int jj = 0; jj != ii; jj++) {
int ord_j = Points_[jj].Oidx();
double dist = FrameDistances_.GetFdist(Points_[ord_i].Fnum(), Points_[ord_j].Fnum());
if (dist < Points_[ord_i].Dist()) {
Points_[ord_i].SetDist( dist );
Points_[ord_j].SetNearestIdx( ord_j );
}
}
}
if (!dvdfile_.empty()) {
CpptrajFile output;
if (output.OpenWrite(dvdfile_)) return 1;
for (Carray::const_iterator point = Points_.begin(); point != Points_.end(); ++point)
output.Printf("%g %g %i\n", point->Density(), point->Dist(), point->NearestIdx()+1);
output.CloseFile();
}
return 0;
}
int Parm_CharmmPsf::WriteParm(FileName const& fname, Topology const& parm) {
// TODO: CMAP etc info
CpptrajFile outfile;
if (outfile.OpenWrite(fname)) return 1;
// Write PSF
outfile.Printf("PSF\n\n");
// Write title
std::string titleOut = parm.ParmName();
titleOut.resize(78);
outfile.Printf("%8i !NTITLE\n* %-78s\n\n", 1, titleOut.c_str());
// Write NATOM section
outfile.Printf("%8i !NATOM\n", parm.Natom());
unsigned int idx = 1;
// Make fake segment ids for now.
char segid[2];
segid[0] = 'A';
segid[1] = '\0';
mprintf("Warning: Assigning single letter segment IDs.\n");
int currentMol = 0;
bool inSolvent = false;
for (Topology::atom_iterator atom = parm.begin(); atom != parm.end(); ++atom, ++idx) {
int resnum = atom->ResNum();
if (atom->MolNum() != currentMol) {
if (!inSolvent) {
inSolvent = parm.Mol(atom->MolNum()).IsSolvent();
currentMol = atom->MolNum();
segid[0]++;
} else
inSolvent = parm.Mol(atom->MolNum()).IsSolvent();
}
// TODO: Print type name for xplor-like PSF
int typeindex = atom->TypeIndex() + 1;
// If type begins with digit, assume charmm numbers were read as
// type. Currently Amber types all begin with letters.
if (isdigit(atom->Type()[0]))
typeindex = convertToInteger( *(atom->Type()) );
// ATOM# SEGID RES# RES ATNAME ATTYPE CHRG MASS (REST OF COLUMNS ARE LIKELY FOR CMAP AND CHEQ)
outfile.Printf("%8i %-4s %-4i %-4s %-4s %4i %14.6G %9g %10i\n", idx, segid,
parm.Res(resnum).OriginalResNum(), parm.Res(resnum).c_str(),
atom->c_str(), typeindex, atom->Charge(),
atom->Mass(), 0);
}
outfile.Printf("\n");
// Write NBOND section
outfile.Printf("%8u !NBOND: bonds\n", parm.Bonds().size() + parm.BondsH().size());
idx = 1;
for (BondArray::const_iterator bond = parm.BondsH().begin();
bond != parm.BondsH().end(); ++bond, ++idx)
{
outfile.Printf("%8i%8i", bond->A1()+1, bond->A2()+1);
if ((idx % 4)==0) outfile.Printf("\n");
}
for (BondArray::const_iterator bond = parm.Bonds().begin();
bond != parm.Bonds().end(); ++bond, ++idx)
{
outfile.Printf("%8i%8i", bond->A1()+1, bond->A2()+1);
if ((idx % 4)==0) outfile.Printf("\n");
}
if ((idx % 4)!=0) outfile.Printf("\n");
outfile.Printf("\n");
// Write NTHETA section
outfile.Printf("%8u !NTHETA: angles\n", parm.Angles().size() + parm.AnglesH().size());
idx = 1;
for (AngleArray::const_iterator ang = parm.AnglesH().begin();
ang != parm.AnglesH().end(); ++ang, ++idx)
{
outfile.Printf("%8i%8i%8i", ang->A1()+1, ang->A2()+1, ang->A3()+1);
if ((idx % 3)==0) outfile.Printf("\n");
}
for (AngleArray::const_iterator ang = parm.Angles().begin();
ang != parm.Angles().end(); ++ang, ++idx)
{
outfile.Printf("%8i%8i%8i", ang->A1()+1, ang->A2()+1, ang->A3()+1);
if ((idx % 3)==0) outfile.Printf("\n");
}
if ((idx % 3)==0) outfile.Printf("\n");
outfile.Printf("\n");
// Write out NPHI section
outfile.Printf("%8u !NPHI: dihedrals\n", parm.Dihedrals().size() + parm.DihedralsH().size());
idx = 1;
for (DihedralArray::const_iterator dih = parm.DihedralsH().begin();
dih != parm.DihedralsH().end(); ++dih, ++idx)
{
outfile.Printf("%8i%8i%8i%8i", dih->A1()+1, dih->A2()+1, dih->A3()+1, dih->A4()+1);
if ((idx % 2)==0) outfile.Printf("\n");
}
for (DihedralArray::const_iterator dih = parm.Dihedrals().begin();
dih != parm.Dihedrals().end(); ++dih, ++idx)
{
outfile.Printf("%8i%8i%8i%8i", dih->A1()+1, dih->A2()+1, dih->A3()+1, dih->A4()+1);
if ((idx % 2)==0) outfile.Printf("\n");
}
if ((idx % 2)==0) outfile.Printf("\n");
outfile.Printf("\n");
outfile.CloseFile();
return 0;
}
// Analysis_Wavelet::Analyze()
Analysis::RetType Analysis_Wavelet::Analyze() {
// Step 1 - Create a matrix that is #atoms rows by #frames - 1 cols,
// where matrix(frame, atom) is the distance that atom has
// travelled from the previous frame.
// TODO: Implement this in Action_Matrix()?
mprintf(" WAVELET:\n");
// First set up atom mask.
if (coords_->Top().SetupIntegerMask( mask_ )) return Analysis::ERR;
mask_.MaskInfo();
int natoms = mask_.Nselected();
int nframes = (int)coords_->Size();
if (natoms < 1 || nframes < 2) {
mprinterr("Error: Not enough frames (%i) or atoms (%i) in '%s'\n",
nframes, natoms, coords_->legend());
return Analysis::ERR;
}
Matrix<double> d_matrix;
mprintf("\t%i frames, %i atoms, distance matrix will require %.2f MB\n",
(double)d_matrix.sizeInBytes(nframes, natoms) / (1024.0*1024.0));
d_matrix.resize(nframes, natoms);
// Get initial frame.
Frame currentFrame, lastFrame;
currentFrame.SetupFrameFromMask( mask_, coords_->Top().Atoms() );
lastFrame = currentFrame;
coords_->GetFrame( 0, lastFrame, mask_ );
// Iterate over frames
for (int frm = 1; frm != nframes; frm++) {
coords_->GetFrame( frm, currentFrame, mask_ );
int idx = frm; // Position in distance matrix; start at column 'frame'
for (int at = 0; at != natoms; at++, idx += nframes)
// Distance of atom in currentFrame from its position in lastFrame.
d_matrix[idx] = sqrt(DIST2_NoImage( currentFrame.XYZ(at), lastFrame.XYZ(at) ));
//lastFrame = currentFrame; // TODO: Re-enable?
}
# ifdef DEBUG_WAVELET
// DEBUG: Write matrix to file.
CpptrajFile dmatrixOut; // DEBUG
dmatrixOut.OpenWrite("dmatrix.dat");
Matrix<double>::iterator mval = d_matrix.begin();
for (int row = 0; row != natoms; row++) {
for (int col = 0; col != nframes; col++)
dmatrixOut.Printf("%g ", *(mval++));
dmatrixOut.Printf("\n");
}
dmatrixOut.CloseFile();
# endif
// Precompute some factors for calculating scaled wavelets.
const double one_over_sqrt_N = 1.0 / sqrt(static_cast<double>( nframes ));
std::vector<int> arrayK( nframes );
arrayK[0] = -1 * (nframes/2);
for (int i = 1; i != nframes; i++)
arrayK[i] = arrayK[i-1] + 1;
# ifdef DEBUG_WAVELET
mprintf("DEBUG: K:");
for (std::vector<int>::const_iterator kval = arrayK.begin(); kval != arrayK.end(); ++kval)
mprintf(" %i", *kval);
mprintf("\n");
# endif
// Step 2 - Get FFT of wavelet for each scale.
PubFFT pubfft;
pubfft.SetupFFTforN( nframes );
mprintf("\tMemory required for scaled wavelet array: %.2f MB\n",
(double)(2 * nframes * nb_ * sizeof(double)) / (1024 * 1024));
typedef std::vector<ComplexArray> WaveletArray;
WaveletArray FFT_of_Scaled_Wavelets;
FFT_of_Scaled_Wavelets.reserve( nb_ );
typedef std::vector<double> Darray;
Darray scaleVector;
scaleVector.reserve( nb_ );
Darray MIN( nb_ * 2 );
for (int iscale = 0; iscale != nb_; iscale++)
{
// Calculate and store scale factor.
scaleVector.push_back( S0_ * pow(2.0, iscale * ds_) );
// Populate MIN array
MIN[iscale ] = (0.00647*pow((correction_*scaleVector.back()),1.41344)+19.7527)*chival_;
MIN[iscale+nb_] = correction_*scaleVector.back();
// Calculate scalved wavelet
ComplexArray scaledWavelet;
switch (wavelet_type_) {
case W_MORLET:
scaledWavelet = F_Morlet(arrayK, scaleVector.back());
break;
case W_PAUL :
scaledWavelet = F_Paul(arrayK, scaleVector.back());
break;
case W_NONE :
return Analysis::ERR; // Sanity check
}
# ifdef DEBUG_WAVELET
PrintComplex("wavelet_before_fft", scaledWavelet);
# endif
// Perform FFT
pubfft.Forward( scaledWavelet );
// Normalize
scaledWavelet.Normalize( one_over_sqrt_N );
# ifdef DEBUG_WAVELET
PrintComplex("wavelet_after_fft", scaledWavelet);
# endif
FFT_of_Scaled_Wavelets.push_back( scaledWavelet );
}
# ifdef DEBUG_WAVELET
mprintf("DEBUG: Scaling factors:");
for (Darray::const_iterator sval = scaleVector.begin(); sval != scaleVector.end(); ++sval)
mprintf(" %g", *sval);
mprintf("\n");
mprintf("DEBUG: MIN:");
for (int i = 0; i != nb_; i++)
mprintf(" %g", MIN[i]);
mprintf("\n");
# endif
// Step 3 - For each atom, calculate the convolution of scaled wavelets
// with rows (atom distance vs frame) via dot product of the
// frequency domains, i.e. Fourier-transformed, followed by an
// inverse FT.
DataSet_MatrixFlt& OUT = static_cast<DataSet_MatrixFlt&>( *output_ );
mprintf("\tMemory required for output matrix: %.2f MB\n",
(double)Matrix<float>::sizeInBytes(nframes, natoms)/(1024.0*1024.0));
OUT.Allocate2D( nframes, natoms ); // Should initialize to zero
Matrix<double> MAX;
mprintf("\tMemory required for Max array: %.2f MB\n",
(double)MAX.sizeInBytes(nframes, natoms)/(1024.0*1024.0));
MAX.resize( nframes, natoms );
Darray magnitude( nframes ); // Scratch space for calculating magnitude across rows
for (int at = 0; at != natoms; at++) {
ComplexArray AtomSignal( nframes ); // Initializes to zero
// Calculate the distance variance for this atom and populate the array.
int midx = at * nframes; // Index into d_matrix
int cidx = 0; // Index into AtomSignal
double d_avg = 0.0;
double d_var = 0.0;
for (int frm = 0; frm != nframes; frm++, cidx += 2, midx++) {
d_avg += d_matrix[midx];
d_var += (d_matrix[midx] * d_matrix[midx]);
AtomSignal[cidx] = d_matrix[midx];
}
d_var = (d_var - ((d_avg * d_avg) / (double)nframes)) / ((double)(nframes - 1));
# ifdef DEBUG_WAVELET
mprintf("VARIANCE: %g\n", d_var);
# endif
double var_norm = 1.0 / d_var;
// Calculate FT of atom signal
pubfft.Forward( AtomSignal );
# ifdef DEBUG_WAVELET
PrintComplex("AtomSignal", AtomSignal);
# endif
// Normalize
AtomSignal.Normalize( one_over_sqrt_N );
// Calculate dot product of atom signal with each scaled FT wavelet
for (int iscale = 0; iscale != nb_; iscale++) {
ComplexArray dot = AtomSignal.TimesComplexConj( FFT_of_Scaled_Wavelets[iscale] );
// Inverse FT of dot product
pubfft.Back( dot );
# ifdef DEBUG_WAVELET
PrintComplex("InverseFT_Dot", dot);
# endif
// Chi-squared testing
midx = at * nframes;
cidx = 0;
for (int frm = 0; frm != nframes; frm++, cidx += 2, midx++) {
magnitude[frm] = (dot[cidx]*dot[cidx] + dot[cidx+1]*dot[cidx+1]) * var_norm;
if (magnitude[frm] < MIN[iscale])
magnitude[frm] = 0.0;
if (magnitude[frm] > MAX[midx]) {
MAX[midx] = magnitude[frm];
//Indices[midx] = iscale
OUT[midx] = (float)(correction_ * scaleVector[iscale]);
}
}
# ifdef DEBUG_WAVELET
mprintf("DEBUG: AbsoluteValue:");
for (Darray::const_iterator dval = magnitude.begin(); dval != magnitude.end(); ++dval)
mprintf(" %g", *dval);
mprintf("\n");
# endif
} // END loop over scales
} // END loop over atoms
# ifdef DEBUG_WAVELET
// DEBUG: Print MAX
CpptrajFile maxmatrixOut; // DEBUG
maxmatrixOut.OpenWrite("maxmatrix.dat");
for (int col = 0; col != nframes; col++) {
for (int row = 0; row != natoms; row++)
maxmatrixOut.Printf("%g ", MAX.element(col, row));
maxmatrixOut.Printf("\n");
}
maxmatrixOut.CloseFile();
# endif
return Analysis::OK;
}
// Cluster_DBSCAN::ComputeKdistMap()
void Cluster_DBSCAN::ComputeKdistMap( Range const& Kvals,
std::vector<int> const& FramesToCluster ) const
{
int pt1_idx, pt2_idx, d_idx, point;
mprintf("\tCalculating Kdist map for %s\n", Kvals.RangeArg());
double* kdist_array; // Store distance of pt1 to every other point.
int nframes = (int)FramesToCluster.size();
// Ensure all Kdist points are within proper range
Range::const_iterator kval;
for (kval = Kvals.begin(); kval != Kvals.end(); ++kval)
if (*kval < 1 || *kval >= nframes) {
mprinterr("Error: Kdist value %i is out of range (1 <= Kdist < %i)\n",
*kval, nframes);
return;
}
int nvals = (int)Kvals.Size();
double** KMAP; // KMAP[i] has the ith nearest point for each point.
KMAP = new double*[ nvals ];
for (int i = 0; i != nvals; i++)
KMAP[i] = new double[ nframes ];
ParallelProgress progress( nframes );
# ifdef _OPENMP
# pragma omp parallel private(pt1_idx, pt2_idx, d_idx, kval, point, kdist_array) firstprivate(progress)
{
progress.SetThread( omp_get_thread_num() );
#endif
kdist_array = new double[ nframes ];
# ifdef _OPENMP
# pragma omp for
# endif
for (pt1_idx = 0; pt1_idx < nframes; pt1_idx++) // X
{
progress.Update( pt1_idx );
point = FramesToCluster[pt1_idx];
d_idx = 0;
// Store distances from pt1 to pt2
for (pt2_idx = 0; pt2_idx != nframes; pt2_idx++)
kdist_array[d_idx++] = FrameDistances_.GetFdist(point, FramesToCluster[pt2_idx]);
// Sort distances; will be smallest to largest
std::sort( kdist_array, kdist_array + nframes );
// Save the distance of specified nearest neighbors to this point.
d_idx = 0;
for (kval = Kvals.begin(); kval != Kvals.end(); ++kval) // Y
KMAP[d_idx++][pt1_idx] = kdist_array[ *kval ];
}
delete[] kdist_array;
# ifdef _OPENMP
} // END omp parallel
# endif
progress.Finish();
// Sort all of the individual kdist plots, smallest to largest.
for (int i = 0; i != nvals; i++)
std::sort(KMAP[i], KMAP[i] + nframes);
// Save in matrix, largest to smallest.
DataSet_MatrixDbl kmatrix;
kmatrix.Allocate2D( FramesToCluster.size(), Kvals.Size() );
for (int y = 0; y != nvals; y++) {
for (int x = nframes - 1; x != -1; x--)
kmatrix.AddElement( KMAP[y][x] );
delete[] KMAP[y];
}
delete[] KMAP;
// Write matrix to file
DataFile outfile;
ArgList outargs("usemap");
outfile.SetupDatafile(k_prefix_ + "Kmatrix.gnu", outargs, debug_);
outfile.AddDataSet( (DataSet*)&kmatrix );
outfile.WriteDataOut();
// Write out the largest and smallest values for each K.
// This means for each value of K the point with the furthest Kth-nearest
// neighbor etc.
CpptrajFile maxfile;
if (maxfile.OpenWrite(k_prefix_ + "Kmatrix.max.dat")) return;
maxfile.Printf("%-12s %12s %12s\n", "#Kval", "MaxD", "MinD");
d_idx = 0;
for (kval = Kvals.begin(); kval != Kvals.end(); ++kval, d_idx++)
maxfile.Printf("%12i %12g %12g\n", *kval, kmatrix.GetElement(0, d_idx),
kmatrix.GetElement(nframes-1, d_idx));
maxfile.CloseFile();
}