#include "readphylipvector.h"
//**********************************************************************************************************************
-vector<string> NMDSCommand::getValidParameters(){
+vector<string> NMDSCommand::setParameters(){
try {
- string Array[] = {"phylip","axes","mindim","maxdim","iters","maxiters","trace","epsilon","outputdir","inputdir"};
- vector<string> myArray (Array, Array+(sizeof(Array)/sizeof(string)));
+ CommandParameter paxes("axes", "InputTypes", "", "", "none", "none", "none",false,false); parameters.push_back(paxes);
+ CommandParameter pphylip("phylip", "InputTypes", "", "", "none", "none", "none",false,true); parameters.push_back(pphylip);
+ CommandParameter pmaxdim("maxdim", "Number", "", "2", "", "", "",false,false); parameters.push_back(pmaxdim);
+ CommandParameter pmindim("mindim", "Number", "", "2", "", "", "",false,false); parameters.push_back(pmindim);
+ CommandParameter piters("iters", "Number", "", "10", "", "", "",false,false); parameters.push_back(piters);
+ CommandParameter pmaxiters("maxiters", "Number", "", "500", "", "", "",false,false); parameters.push_back(pmaxiters);
+ CommandParameter pepsilon("epsilon", "Number", "", "0.000000000001", "", "", "",false,false); parameters.push_back(pepsilon);
+ CommandParameter pinputdir("inputdir", "String", "", "", "", "", "",false,false); parameters.push_back(pinputdir);
+ CommandParameter poutputdir("outputdir", "String", "", "", "", "", "",false,false); parameters.push_back(poutputdir);
+
+ vector<string> myArray;
+ for (int i = 0; i < parameters.size(); i++) { myArray.push_back(parameters[i].name); }
return myArray;
}
catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "getValidParameters");
- exit(1);
- }
-}
-//**********************************************************************************************************************
-NMDSCommand::NMDSCommand(){
- try {
- abort = true;
- //initialize outputTypes
- vector<string> tempOutNames;
- outputTypes["nmds"] = tempOutNames;
- outputTypes["stress"] = tempOutNames;
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "NMDSCommand");
+ m->errorOut(e, "NMDSCommand", "setParameters");
exit(1);
}
}
//**********************************************************************************************************************
-vector<string> NMDSCommand::getRequiredParameters(){
+string NMDSCommand::getHelpString(){
try {
- string Array[] = {"phylip"};
- vector<string> myArray (Array, Array+(sizeof(Array)/sizeof(string)));
- return myArray;
+ string helpString = "";
+ helpString += "The nmds command is modelled after the nmds code written in R by Sarah Goslee, using Non-metric multidimensional scaling function using the majorization algorithm from Borg & Groenen 1997, Modern Multidimensional Scaling.\n";
+ helpString += "The nmds command parameters are phylip, axes, mindim, maxdim, maxiters, iters and epsilon.\n";
+ helpString += "The phylip parameter allows you to enter your distance file.\n";
+ helpString += "The axes parameter allows you to enter a file containing a starting configuration.\n";
+ helpString += "The maxdim parameter allows you to select the maximum dimensions to use. Default=2\n";
+ helpString += "The mindim parameter allows you to select the minimum dimensions to use. Default=2\n";
+ helpString += "The maxiters parameter allows you to select the maximum number of iters to try with each random configuration. Default=500\n";
+ helpString += "The iters parameter allows you to select the number of random configuration to try. Default=10\n";
+ helpString += "The epsilon parameter allows you to select set an acceptable stopping point. Default=1e-12.\n";
+ helpString += "Example nmds(phylip=yourDistanceFile).\n";
+ helpString += "Note: No spaces between parameter labels (i.e. phylip), '=' and parameters (i.e.yourDistanceFile).\n";
+ return helpString;
}
catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "getRequiredParameters");
+ m->errorOut(e, "NMDSCommand", "getHelpString");
exit(1);
}
}
//**********************************************************************************************************************
-vector<string> NMDSCommand::getRequiredFiles(){
+NMDSCommand::NMDSCommand(){
try {
- vector<string> myArray;
- return myArray;
+ abort = true; calledHelp = true;
+ setParameters();
+ vector<string> tempOutNames;
+ outputTypes["nmds"] = tempOutNames;
+ outputTypes["stress"] = tempOutNames;
+ outputTypes["iters"] = tempOutNames;
}
catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "getRequiredFiles");
+ m->errorOut(e, "NMDSCommand", "NMDSCommand");
exit(1);
}
}
NMDSCommand::NMDSCommand(string option) {
try {
- abort = false;
+ abort = false; calledHelp = false;
//allow user to run help
- if(option == "help") { help(); abort = true; }
+ if(option == "help") { help(); abort = true; calledHelp = true; }
else {
- //valid paramters for this command
- string Array[] = {"phylip","axes","mindim","maxdim","iters","maxiters","trace","epsilon","outputdir", "inputdir"};
- vector<string> myArray (Array, Array+(sizeof(Array)/sizeof(string)));
+ vector<string> myArray = setParameters();
OptionParser parser(option);
map<string, string> parameters = parser. getParameters();
//initialize outputTypes
vector<string> tempOutNames;
outputTypes["nmds"] = tempOutNames;
+ outputTypes["iters"] = tempOutNames;
outputTypes["stress"] = tempOutNames;
//required parameters
phylipfile = validParameter.validFile(parameters, "phylip", true);
if (phylipfile == "not open") { phylipfile = ""; abort = true; }
- else if (phylipfile == "not found") { phylipfile = ""; m->mothurOut("You must provide a distance file before running the nmds command."); m->mothurOutEndLine(); abort = true; }
+ else if (phylipfile == "not found") {
+ //if there is a current phylip file, use it
+ phylipfile = m->getPhylipFile();
+ if (phylipfile != "") { m->mothurOut("Using " + phylipfile + " as input file for the phylip parameter."); m->mothurOutEndLine(); }
+ else { m->mothurOut("You have no current phylip file and the phylip parameter is required."); m->mothurOutEndLine(); abort = true; }
+ }
axesfile = validParameter.validFile(parameters, "axes", true);
if (axesfile == "not open") { axesfile = ""; abort = true; }
outputDir += m->hasPath(phylipfile); //if user entered a file with a path then preserve it
}
- string temp = validParameter.validFile(parameters, "mindim", false); if (temp == "not found") { temp = "1"; }
+ string temp = validParameter.validFile(parameters, "mindim", false); if (temp == "not found") { temp = "2"; }
convert(temp, mindim);
temp = validParameter.validFile(parameters, "maxiters", false); if (temp == "not found") { temp = "500"; }
temp = validParameter.validFile(parameters, "epsilon", false); if (temp == "not found") { temp = "0.000000000001"; }
convert(temp, epsilon);
- temp = validParameter.validFile(parameters, "trace", false); if (temp == "not found") { temp = "F"; }
- trace = m->isTrue(temp);
-
if (mindim < 1) { m->mothurOut("mindim must be at least 1."); m->mothurOutEndLine(); abort = true; }
if (maxdim < mindim) { m->mothurOut("maxdim must be greater than mindim."); m->mothurOutEndLine(); abort = true; }
}
}
}
//**********************************************************************************************************************
-void NMDSCommand::help(){
- try {
- m->mothurOut("The nmds command is modelled after the nmds code written in R by Sarah Goslee, using Non-metric multidimensional scaling function using the majorization algorithm from Borg & Groenen 1997, Modern Multidimensional Scaling."); m->mothurOutEndLine();
- m->mothurOut("The nmds command parameters are phylip, axes, mindim, maxdim, maxiters, iters, epsilon and trace."); m->mothurOutEndLine();
- m->mothurOut("The phylip parameter allows you to enter your distance file."); m->mothurOutEndLine();
- m->mothurOut("The axes parameter allows you to enter a file containing a starting configuration."); m->mothurOutEndLine();
- m->mothurOut("The maxdim parameter allows you to select how maximum dimensions to use. Default=2"); m->mothurOutEndLine();
- m->mothurOut("The mindim parameter allows you to select how minimum dimensions to use. Default=1"); m->mothurOutEndLine();
- m->mothurOut("The maxiters parameter allows you to select the maximum number of iters to try with each random configuration. Default=500"); m->mothurOutEndLine();
- m->mothurOut("The iters parameter allows you to select the number of random configuration to try. Default=10"); m->mothurOutEndLine();
- m->mothurOut("The epsilon parameter allows you to select set an acceptable stopping point. Default=1e-12."); m->mothurOutEndLine();
- m->mothurOut("The trace parameter allows you to see the output after each iter. Default=F"); m->mothurOutEndLine();
- m->mothurOut("Example nmds(phylip=yourDistanceFile).\n");
- m->mothurOut("Note: No spaces between parameter labels (i.e. phylip), '=' and parameters (i.e.yourDistanceFile).\n\n");
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "help");
- exit(1);
- }
-}
-//**********************************************************************************************************************
-NMDSCommand::~NMDSCommand(){}
-//**********************************************************************************************************************
int NMDSCommand::execute(){
try {
- if (abort == true) { return 0; }
+ if (abort == true) { if (calledHelp) { return 0; } return 2; }
cout.setf(ios::fixed, ios::floatfield);
cout.setf(ios::showpoint);
vector< vector<double> > axes;
if (axesfile != "") { axes = readAxes(names); }
+ string outputFileName = outputDir + m->getRootName(m->getSimpleName(phylipfile)) + "nmds.iters";
+ string stressFileName = outputDir + m->getRootName(m->getSimpleName(phylipfile)) + "nmds.stress";
+ outputNames.push_back(outputFileName); outputTypes["iters"].push_back(outputFileName);
+ outputNames.push_back(stressFileName); outputTypes["stress"].push_back(stressFileName);
+
+ ofstream out, out2;
+ m->openOutputFile(outputFileName, out);
+ m->openOutputFile(stressFileName, out2);
+
+ out2.setf(ios::fixed, ios::floatfield);
+ out2.setf(ios::showpoint);
+ out.setf(ios::fixed, ios::floatfield);
+ out.setf(ios::showpoint);
+
+ out2 << "Dimension\tIter\tStress\tRsq" << endl;
+
+ double bestStress = 10000000;
+ double bestR2 = 10000000;
+ vector< vector<double> > bestConfig;
+ int bestDim = 0;
+
for (int i = mindim; i <= maxdim; i++) {
m->mothurOut("Processing Dimension: " + toString(i)); m->mothurOutEndLine();
- string outputFileName = outputDir + m->getRootName(m->getSimpleName(phylipfile)) + "dim" + toString(i) + ".nmds";
- string stressFileName = outputDir + m->getRootName(m->getSimpleName(phylipfile)) + "dim" + toString(i) + ".stress.nmds";
- outputNames.push_back(outputFileName); outputTypes["nmds"].push_back(outputFileName);
- outputNames.push_back(stressFileName); outputTypes["stress"].push_back(stressFileName);
-
- ofstream out, out2;
- m->openOutputFile(outputFileName, out);
- m->openOutputFile(stressFileName, out2);
-
- out2.setf(ios::fixed, ios::floatfield);
- out2.setf(ios::showpoint);
- out.setf(ios::fixed, ios::floatfield);
- out.setf(ios::showpoint);
-
- out2 << "Iter\tStress\tCorr" << endl;
-
for (int j = 0; j < iters; j++) {
- if (trace) { m->mothurOut(toString(j+1)); m->mothurOutEndLine(); }
+ m->mothurOut(toString(j+1)); m->mothurOutEndLine();
//get configuration - either randomly generate or resize to this dimension
vector< vector<double> > thisConfig;
if (m->control_pressed) { out.close(); out2.close(); for (int k = 0; k < outputNames.size(); k++) { remove(outputNames[k].c_str()); } return 0; }
//calc correlation between original distances and euclidean distances from this config
- double corr = linearCalc.calcPearson(matrix, newEuclid);
- corr *= corr;
+ double rsquared = linearCalc.calcPearson(newEuclid, matrix);
+ rsquared *= rsquared;
if (m->control_pressed) { out.close(); out2.close(); for (int k = 0; k < outputNames.size(); k++) { remove(outputNames[k].c_str()); } return 0; }
//output results
out << "Config" << (j+1) << '\t';
- for (int k = 0; k < i; k++) { out << "X" << (k+1) << '\t'; }
+ for (int k = 0; k < i; k++) { out << "axis" << (k+1) << '\t'; }
out << endl;
- out2 << (j+1) << '\t' << stress << '\t' << corr << endl;
+ out2 << i << '\t' << (j+1) << '\t' << stress << '\t' << rsquared << endl;
output(endConfig, names, out);
+ //save best
+ if (stress < bestStress) {
+ bestDim = i;
+ bestStress = stress;
+ bestR2 = rsquared;
+ bestConfig = endConfig;
+ }
+
if (m->control_pressed) { out.close(); out2.close(); for (int k = 0; k < outputNames.size(); k++) { remove(outputNames[k].c_str()); } return 0; }
-
}
-
- out.close(); out2.close();
}
+ out.close(); out2.close();
+
+ //output best config
+ string BestFileName = outputDir + m->getRootName(m->getSimpleName(phylipfile)) + "nmds.axes";
+ outputNames.push_back(BestFileName); outputTypes["nmds"].push_back(BestFileName);
+
+ m->mothurOut("\nNumber of dimensions:\t" + toString(bestDim) + "\n");
+ m->mothurOut("Lowest stress :\t" + toString(bestStress) + "\n");
+ m->mothurOut("R-squared for configuration:\t" + toString(bestR2) + "\n");
+
+ ofstream outBest;
+ m->openOutputFile(BestFileName, outBest);
+ outBest.setf(ios::fixed, ios::floatfield);
+ outBest.setf(ios::showpoint);
+
+ outBest << '\t';
+ for (int k = 0; k < bestConfig.size(); k++) { outBest << "axis" << (k+1) << '\t'; }
+ outBest << endl;
+
+ output(bestConfig, names, outBest);
+
+ outBest.close();
+
if (m->control_pressed) { for (int i = 0; i < outputNames.size(); i++) { remove(outputNames[i].c_str()); } return 0; }
m->mothurOutEndLine();
double stress2 = calculateStress(matrix, euclid);
stress1 = stress2 + 1.0 + epsilon;
- if (trace) { m->mothurOutEndLine(); m->mothurOut("Iter\tStress"); m->mothurOutEndLine(); }
-
int count = 0;
while ((count < maxIters) && (abs(stress1 - stress2) > epsilon)) {
count++;
euclid = linearCalc.calculateEuclidianDistance(newConfig);
stress2 = calculateStress(matrix, euclid);
-
- if (trace) { m->mothurOut(count + "\t" + toString(stress1)); m->mothurOutEndLine(); }
-
}
return newConfig;
exit(1);
}
}
-/**********************************************************************************************************************
- vector< vector<double> > NMDSCommand::calculateStressGradientVector(vector<seqDist>& eDists, vector<seqDist>& D, double rawStress, double stress, vector< vector<double> >& axes) {
- try {
- vector< vector<double> > gradient; gradient.resize(dimension);
- for (int i = 0; i < gradient.size(); i++) { gradient[i].resize(axes[0].size(), 0.0); }
-
- double sumDij = 0.0;
- for (int i = 0; i < eDists.size(); i++) { sumDij += (eDists[i].dist * eDists[i].dist); }
-
- for (int i = 0; i < eDists.size(); i++) {
-
- for (int j = 0; j < dimension; j++) {
-
- if (m->control_pressed) { return gradient; }
-
- double firstTerm1 = (stress / rawStress) * (eDists[i].dist - D[i].dist);
- double firstTerm2 = (stress / sumDij) * eDists[i].dist;
- double firstTerm = firstTerm1 - firstTerm2;
-
- float r = (dimension-1.0);
- double temp = 1.0 / (pow(eDists[i].dist, r));
- float absTemp = abs(axes[j][eDists[i].seq1] - axes[j][eDists[i].seq2]);
- double secondTerm = pow(absTemp, r) * temp;
-
- double sigNum = 1.0;
- if ((axes[j][eDists[i].seq1] - axes[j][eDists[i].seq2]) == 0) { sigNum = 0.0; }
- else if ((axes[j][eDists[i].seq1] - axes[j][eDists[i].seq2]) < 0) { sigNum = -1.0; }
-
- double results = (firstTerm * secondTerm * sigNum);
- cout << i << '\t' << j << '\t' << "results = " << results << endl;
- gradient[j][eDists[i].seq1] += results;
- gradient[j][eDists[i].seq2] -= results;
- }
- }
-
- return gradient;
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "calculateStressGradientVector");
- exit(1);
- }
- }
- //**********************************************************************************************************************
- double NMDSCommand::calculateMagnitude(vector< vector<double> >& gradient) {
- try {
- double magnitude = 0.0;
-
- double sum = 0.0;
- for (int i = 0; i < gradient.size(); i++) {
- for (int j = 0; j < gradient[i].size(); j++) {
- sum += (gradient[i][j] * gradient[i][j]);
- }
- }
-
- magnitude = sqrt(((1.0/(float)gradient[0].size()) * sum));
-
- return magnitude;
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "calculateMagnitude");
- exit(1);
- }
- }
- //**********************************************************************************************************************
- //described in Kruskal paper page 121 + 122
- double NMDSCommand::calculateStep(vector< vector<double> >& prevGrad, vector< vector<double> >& grad, vector<double>& prevStress) {
- try {
- double newStep = step;
-
- //calc the cos theta
- double sumNum = 0.0;
- double sumDenom1 = 0.0;
- double sumDenom2 = 0.0;
- for (int i = 0; i < prevGrad.size(); i++) {
- for (int j = 0; j < prevGrad[i].size(); j++) {
- sumDenom1 += (grad[i][j] * grad[i][j]);
- sumDenom2 += (prevGrad[i][j] * prevGrad[i][j]);
- sumNum += (grad[i][j] * prevGrad[i][j]);
- }
- }
-
- double cosTheta = sumNum / (sqrt(sumDenom1) * sqrt(sumDenom2));
- cosTheta *= cosTheta;
-
- //calc angle factor
- double angle = pow(4.0, cosTheta);
-
- //calc 5 step ratio
- double currentStress = prevStress[prevStress.size()-1];
- double lastStress = prevStress[0];
- if (prevStress.size() > 1) { lastStress = prevStress[prevStress.size()-2]; }
- double fivePrevStress = prevStress[0];
- if (prevStress.size() > 5) { fivePrevStress = prevStress[prevStress.size()-6]; }
-
- double fiveStepRatio = min(1.0, (currentStress / fivePrevStress));
-
- //calc relaxation factor
- double relaxation = 1.3 / (1.0 + pow(fiveStepRatio, 5.0));
-
- //calc good luck factor
- double goodLuck = min(1.0, (currentStress / lastStress));
-
- //calc newStep
- //cout << "\ncos = " << cosTheta << " step = " << step << " angle = " << angle << " relaxation = " << relaxation << " goodluck = " << goodLuck << endl;
- newStep = step * angle * relaxation * goodLuck;
-
- return newStep;
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "calculateStep");
- exit(1);
- }
- }
- //**********************************************************************************************************************
- vector< vector<double> > NMDSCommand::calculateNewConfiguration(double magnitude, vector< vector<double> >& axes, vector< vector<double> >& gradient) {
- try {
-
- vector< vector<double> > newAxes = axes;
-
- for (int i = 0; i < newAxes.size(); i++) {
-
- if (m->control_pressed) { return newAxes; }
-
- for (int j = 0; j < newAxes[i].size(); j++) {
- newAxes[i][j] = axes[i][j] + ((step / magnitude) * gradient[i][j]);
- }
- }
-
- return newAxes;
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "calculateNewConfiguration");
- exit(1);
- }
- }*/
-/**********************************************************************************************************************
- //adjust eDists so that it creates monotonically increasing series of succesive values that increase or stay the same, but never decrease
- vector<seqDist> NMDSCommand::satisfyMonotonicity(vector<seqDist> eDists, vector<int> partitions) {
- try {
-
- //find averages of each partitions
- vector<double> sums; sums.resize(partitions.size(), 0.0);
- vector<int> sizes; sizes.resize(partitions.size(), 0);
-
- for (int i = 0; i < partitions.size(); i++) {
- //i is not the last one
- int start = partitions[i];
- int end;
- if (i != (partitions.size()-1)) { end = partitions[i+1]; }
- else{ end = eDists.size(); }
-
- for (int j = start; j < end; j++) { sums[i] += eDists[j].dist; }
-
- sizes[i] = (end - start);
- }
-
-
- vector<seqDist> D = eDists;
-
- //i represents the "active block"
- int i = 0;
- while (i < partitions.size()) {
-
- if (m->control_pressed) { return D; }
-
- bool upActive = true;
- bool upSatisfied = false;
- bool downSatisfied = false;
-
- //while we are not done with this block
- while ((!upSatisfied) || (!downSatisfied)) {
-
- if (upActive) {
-
- //are we are upSatisfied? - is the average of the next block greater than mine?
- if (i != (partitions.size()-1)) { //if we are the last guy then we are upsatisfied
- if ((sums[i+1]/(float)sizes[i+1]) >= (sums[i]/(float)sizes[i])) {
- upSatisfied = true;
- upActive = false;
- }else {
- //find new weighted average
- double newSum = sums[i] + sums[i+1];
-
- //merge blocks - putting everything in i
- sums[i] = newSum;
- sizes[i] += sizes[i+1];
- partitions[i] = partitions[i+1];
-
- sums.erase(sums.begin()+(i+1));
- sizes.erase(sizes.begin()+(i+1));
- partitions.erase(partitions.begin()+(i+1));
-
- upActive = false;
- }
- }else { upSatisfied = true; upActive = false; }
-
- }else { //downActive
-
- //are we are DownSatisfied? - is the average of the previous block less than mine?
- if (i != 0) { //if we are the first guy then we are downSatisfied
- if ((sums[i-1]/(float)sizes[i-1]) <= (sums[i]/(float)sizes[i])) {
- downSatisfied = true;
- upActive = true;
- }else {
- //find new weighted average
- double newSum = sums[i] + sums[i-1];;
-
- //merge blocks - putting everything in i-1
- sums[i-1] = newSum;
- sizes[i-1] += sizes[i];
-
- sums.erase(sums.begin()+i);
- sizes.erase(sizes.begin()+i);
- partitions.erase(partitions.begin()+i);
- i--;
-
- upActive = true;
- }
- }else { downSatisfied = true; upActive = true; }
- }
- }
-
- i++; // go to next block
- }
-
- //sanity check - for rounding errors
- vector<double> averages; averages.resize(sums.size(), 0.0);
- for (int i = 0; i < sums.size(); i++) { averages[i] = sums[i] / (float) sizes[i]; }
- for (int i = 0; i < averages.size(); i++) { if (averages[i+1] < averages[i]) { averages[i+1] = averages[i]; } }
-
- //fill D
- int placeHolder = 0;
- for (int i = 0; i < averages.size(); i++) {
- for (int j = 0; j < sizes[i]; j++) {
- D[placeHolder].dist = averages[i];
- placeHolder++;
- }
- }
-
- return D;
- }
- catch(exception& e) {
- m->errorOut(e, "NMDSCommand", "satisfyMonotonicity");
- exit(1);
- }
- }*/
+/**********************************************************************************************************************/
-//**********************************************************************************************************************