private static void initFrequencys(Model parent, SuperFrequencyTrace frequencys) {
// assume frequecys is set to end.
PopulationSizeModel[] sizes = parent.getPopulationSizeModels();
double total = 0;
for (int d = 0; d < sizes.length; d++) {
total += sizes[d].populationSize(0); // assuming homogenous time
// model.
}
double pick = RandomGenerator.getRandom().nextDouble() * total;
int d = 0;
total = 0;
for (int i = 0; i < sizes.length; i++) {
total += sizes[d].populationSize(0);
if (total >= pick) {
d = i;
break;
}
}
int choosen = parent.getModelHistory().getSeedDeme();
if (choosen > -1) d = choosen;
double[] freqs = new double[sizes.length]; // demeCount;
freqs[d] =
1.0 / (parent.getPopulationSizeModels()[d].populationSize(parent.getEndTime())); // FIXME
frequencys.setFrequencys(freqs);
// System.out.println("INITF:"+Arrays.toString(freqs));
}
/**
* The basic simulator. Its getting more rewrites than you can shake a stick at. What is next? Well
* we need to simplify the dam frequency trace stuff. Its too fragile and is using too much memory.
* Push back to a single dimension array and do index arithmetic is a good start. However still have
* lots of "conditions". For example we have stopping conditions. There is also restart conditions.
* Restart conditions could go over model boundaries. So this needs to go outside this classes
* scope. finally there are time shift Conditions when we use fixation time for finishing a sweep.
* Also there is the situation where we may want the sweep to continue after such a condition is
* meet. We also have a contract to add the correct start/stop selection events for this coalsecent
* pass.
*
* <p>So even with the move to a simple single dim array that will save memory and avoid cache
* pollution. I don't know how fix the other problems in a clean fashion. Consider that many loci
* code will be added soon! Then there will be haplotypes and alleles.
*
* <p>For now lets consider than time information is *not* in frequency trace but in the model.
*
* <p>So now we have a "iteration" of selection. Check for boundary conditions. Resets are bubbled
* up the call stack and must deal with removing any Cevent.
*
* @author bob
*/
public class DefaultSelectionSimulator implements SelectionSimulator {
private Binomial binomial = RandomGenerator.getBinomial();
// private Random random=RandomGenerator.getRandom();
private int ndeme;
private PopulationSizeModel[] sizes; // =model.getPopulationSizeModels();
private double[] totalMRates; // =model.getTotalMigrationRates();
private int[][] migrationDirections; // =model.getMigrationDirectionsByDeme();
private double[][] migrationRates; // =model.getMigrationRatesByDeme();
private double mu; // =modelHistory.getForwardAlleleMutationRate();
private double nu; // =modelHistory.getBackAlleleMutationRate();
private double[] afterSelectionA; // =new double[ndeme];
private double[] afterSelectiona; // =new double[ndeme];
public DefaultSelectionSimulator() {
// TODO Auto-generated constructor stub
}
@Override
public List<ModelEvent> init(ModelHistroy modelHistory) {
return new ArrayList<ModelEvent>();
}
// private Binomial binomial=RandomGenerator.getBinomial();
// @Override
// public List<ModelEvent> forwardSimulator(Model model, ModelHistroy modelHistory,
// SelectionStrengthModel[] ssm, SuperFrequencyTrace frequencys) {
// // first init.
//
// if (model.getEndTime() == Long.MAX_VALUE) {
// // for now no initial frequencys are set.
// initFrequencys(model, frequencys);
// }
// List<ModelEvent> events =new ArrayList<ModelEvent>();
// int ndeme =model.getDemeCount();
// // int end=frequencys[0][0].length - 1;
// // int currentGen =frequencys[0][0].length - 1;// we count down: Always.
// // int genCount=0;
//
// double[] afterSelectionA =new double[ndeme];
// double[] afterSelectiona =new double[ndeme];
//
// double[] totalMRates =model.getTotalMigrationRates();
// int[][] migrationDirections =model.getMigrationDirectionsByDeme();
// double[][] migrationRates =model.getMigrationRatesByDeme();
// double mu =modelHistory.getForwardAlleleMutationRate();
// double nu =modelHistory.getBackAlleleMutationRate();
// PopulationSizeModel[] sizes =model.getPopulationSizeModels();
//
// boolean zeroFlag =true;
// boolean fixFlag =true;
// boolean addedSelectionStart =false;
// boolean localTimeRescalling =model.getEndTime() == Long.MAX_VALUE;
//
// FrequencyCondition stopping =modelHistory.getTimedCondition();
// // System.out.println("Stopping:"+stopping);
// frequencys.reset();
// // frequencys.setIteratorToEnd();
// while (false && localTimeRescalling) {
//
// // System.out.println("In Loop:"+frequencys.getTime());
// // first get all the selection amounts in each deme before
// // migration.
//
// // note this with be Integer.MAX_VALUE for fixation time invariant boundry conditions
// int time =frequencys.getTime();
// double[][] thisFreq =frequencys.current();
// double[][] nextFreq =frequencys.previous(localTimeRescalling);
// for (int d =0; d < ndeme; d++) {
// if (sizes[d] == PopulationSizeModel.NULL_POPULATION)
// continue;
// double sAA =ssm[d].getStrength(1, 1, time);
// double saA =ssm[d].getStrength(0, 1, time);
// double saa =ssm[d].getStrength(0, 0, time);
//
// double x =thisFreq[d][1];// frequencys[d][1][currentGen];// note the 2 allele
// // assumption
// double ax =1.0 - x;
//
// if (x > 0)
// zeroFlag =false;
// if (x < 1.0)
// fixFlag =false;
//
// afterSelectionA[d] =x * (1 + ax * saA + x * sAA);
// afterSelectiona[d] =ax * (1 + x * saA + ax * saa);
// }
//
// // now we deal with migration and update the next generations
// // frequencys.
//
// for (int d =0; d < ndeme; d++) {
// if (sizes[d] == PopulationSizeModel.NULL_POPULATION) {
// nextFreq[d][1] =0;// frequencys[d][1][nextGen] =0;
// nextFreq[d][0] =0;// frequencys[d][0][nextGen] =0;
// continue;
// }
// // first self migration.
// double rate =1 - totalMRates[d];
// double nA =rate * afterSelectionA[d];
// double na =rate * afterSelectiona[d];
//
// int[] directions =migrationDirections[d];
// double[] rates =migrationRates[d];
// for (int count =0; count < directions.length; count++) {
// int j =directions[count];
// rate =rates[count];
// nA +=rate * afterSelectionA[j];
// na +=rate * afterSelectiona[j];
// }
// // now put it together for xp
// double xp =(nA * (1 - nu) + mu * na) / (nA + na);
// // we keep N in the sample for now...
// int N =(int) sizes[d].populationSize(time) * 2;//
// double f =(double) binomial.generateBinomial(N, xp) / N;
// // update
// nextFreq[d][1] =f;
// nextFreq[d][0] =1 - f;
// // System.out.println(f+" "+selectionStrength[d].getStrength(1,
// // 1, time));
// // TIME is wrong because we may shift time...
//
// }
//
// // hard to read and not clear at all IMO.
// // better ideas welcome
// if (localTimeRescalling) {
// if (zeroFlag) {
// // reset
// // System.out.println("RESET");
// frequencys.setIteratorToEnd();
// } else if (fixFlag) {
// // we shift time and break
// // first we want indexOffset to point to currentGen
// frequencys.shiftAndSet((int) model.getStartTime(), (int) model.getStartTime());
// events.add(new SelectionEndTimeEvent(frequencys.getEndTime()));
// events.add(new SelectionStartEvent(frequencys.getStartTime(), null));
// addedSelectionStart =true;
// // do we have enough size?
// localTimeRescalling =false;
// // System.out.println("IffixFlagTrueRescalling:"+time+"\t"+frequencys.getStartTime());
// break;
// } else if (stopping != null && stopping.isMeet(nextFreq)) {
// frequencys.shiftAndSet((int) stopping.getTime(), (int) model.getStartTime());
// events.add(new SelectionEndTimeEvent(frequencys.getEndTime()));
// // do we have enough size?
// localTimeRescalling =false;
// // System.out.println("OtherStopping:"+time);
// }
// } else if (fixFlag) {
// // copy fixed to end and break
// // System.out.println("MakeStart:"+time);
// frequencys.shiftAndSet(frequencys.getTime(), frequencys.getTime());
// // System.out.println("IffixFlagTrue:"+time);
// events.add(new SelectionStartEvent(time, null));
// addedSelectionStart =true; // dirty hack to stop adding a SelectionStart event after
// // loop exit
// // frequencys.setStart();
// // for (int d =0; d < ndeme; d++) {
// // Arrays.fill(frequencys[d][0], 0, currentGen, 0.0);
// // Arrays.fill(frequencys[d][1], 0, currentGen, 1.0);
// // }
//
// break;
// }
// fixFlag =true;
// zeroFlag =true;
// //
// System.out.println("Iterate:"+toString(frequencys.current())+"\tTime:"+time+"\tft:"+frequencys.getTime());
// }
// // if we
// if (!addedSelectionStart) {
//
// // need to start the selection phase when we are not fixed.
// double[] freq =new double[ndeme];
// double[][] local =frequencys.current();
// for (int i =0; i < freq.length; i++) {
// freq[i] =local[i][1];
// }
// // System.out.println("notFixed
// "+Arrays.toString(freq)+"\t"+frequencys.getTime()+"\t"+model.getStartTime()+"\n\t"+events);
// events.add(new SelectionStartEvent(frequencys.getTime(), freq));
// }
// // System.out.println("DoneSelection");
// // so we are *almost* done...
// return events;
//
// }
public List<ModelEvent> forwardSimulator(
Model model,
ModelHistroy modelHistory,
SelectionStrengthModel[] ssm,
SuperFrequencyTrace frequencys) {
// first init.
// System.err.println("Did we even call me?");
// if (model.getEndTime() == Long.MAX_VALUE) {
// // for now no initial frequencys are set.
// initFrequencys(model, frequencys);
// }
List<ModelEvent> events = new ArrayList<ModelEvent>();
ndeme = model.getDemeCount();
// int end=frequencys[0][0].length - 1;
// int currentGen =frequencys[0][0].length - 1;// we count down: Always.
// int genCount=0;
afterSelectionA = new double[ndeme];
afterSelectiona = new double[ndeme];
totalMRates = model.getTotalMigrationRates();
migrationDirections = model.getMigrationDirectionsByDeme();
migrationRates = model.getMigrationRatesByDeme();
mu = modelHistory.getForwardAlleleMutationRate();
nu = modelHistory.getBackAlleleMutationRate();
sizes = model.getPopulationSizeModels();
FrequencyCondition stopping = modelHistory.getTimedCondition();
RestartCondition restartCondition = modelHistory.getRestartCondtion();
// System.err.println("Stop:"+stopping+"\tstart:"+restartCondition);
// boolean zeroFlag =true;
// boolean fixFlag =true;
// so we have 2 basic "modes" Condtion on fixation in time homegenous mode. Start with
// defined conditions.
// in the first case time is not important and we can assume time always ==
// model.getStartTime();
// in the second case we are bounded on the interval start time to end time. We assume that
// our lovely
// frequency trace data struct is allocated in such a way that this will work in both cases.
// in both cases we have just one important difference. We don't add a selection end event
// till we know about the
// selelction startEvent for the first case. Otherwise we do everything quite linear like.
// for now lets just fork to 2 different methods
// right here....
frequencys.reset();
boolean fixationConidtion = model.getEndTime() == Long.MAX_VALUE;
if (fixationConidtion) {
// System.err.println("We are doing it baby");
double time = model.getStartTime(); // we use this time for the whole thing. Time
// invaraint right!
frequencys.setIndexMostPastward(); // start at the very back.
initFrequencys(model, frequencys);
double[] thisGen = frequencys.getFrequencys(null);
double[] nextGen = new double[thisGen.length];
while (true) {
// System.err.println("Oh yea!"+Arrays.toString(thisGen));
selectionSimulationStep(thisGen, nextGen, ssm, time);
if (restartCondition.isMeet(nextGen, ndeme)) {
// System.out.println("Restart?");
return null; // bubble up the call stack to where we need it.
}
// copy data back.
frequencys.moveForward();
frequencys.setFrequencys(nextGen);
// check if we are done.
// System.err.println("Stopping?"+stopping.isMeet(nextGen));
if (stopping.isMeet(nextGen)) {
// System.err.println("We Stoped Doing it:"+Arrays.toString(nextGen)+"\t"+stopping);
// shift and stop
frequencys.setCurrentIndexAsStart(); // good luck with that!
// add events
// System.err.println("TRACE:"+frequencys);
events.add(new SelectionEndTimeEvent(frequencys.getTimeMostPastward()));
events.add(
new SelectionStartEvent(stopping.getTime(), nextGen)); // well we already new That!
frequencys.setEndTime();
// System.err.println(events+"\t"+frequencys.getEndTime()+"\tPastward:"+frequencys.getTimeMostPastward());
return events; // get me out here.
}
if (restartCondition.isMeet(nextGen, ndeme)) {
frequencys.setIndexMostPastward(); // start at the very back.
initFrequencys(model, frequencys);
frequencys.getFrequencys(thisGen);
// System.err.println("START AGAIN:"+Arrays.toString(nextGen)+"\t"+stopping);
continue;
}
double[] tmp = thisGen;
thisGen = nextGen; // no need for a copy
nextGen = tmp; // copied over. We hope!
}
}
// so now we have a much simpler task of starting and finishing things at fixed times.
// first we assume that inital conditions are specified and already done.
frequencys.setIndexMostPastward();
long time = frequencys.getIndexTime();
assert time == model.getEndTime() : time + "\t" + model.getEndTime();
double[] thisGen = frequencys.getFrequencys(null);
double[] nextGen = new double[thisGen.length];
// System.out.println("StartLoop 2 "+restartCondition.isMeet(thisGen, ndeme));
while (frequencys.hasMoreForward()) {
selectionSimulationStep(thisGen, nextGen, ssm, time);
frequencys.moveForward();
frequencys.setFrequencys(nextGen);
time = frequencys.getIndexTime();
// System.out.println("restart.."+restartCondition.isMeet(nextGen,
// ndeme)+"\t"+Arrays.toString(nextGen)+"\t"+ndeme+"\ttime:"+time);
if (restartCondition != null && restartCondition.isMeet(nextGen, ndeme)) {
return null; // bubble the restart up the stack.
}
// System.err.println("Stopping?"+stopping.isMeet(nextGen));
if (stopping != null && stopping.isMeet(nextGen)) {
events.add(new SelectionStartEvent(time, thisGen));
// System.out.println("Stop:"+events+"\t"+Arrays.toString(nextGen)+"\t"+time);
frequencys.setEndTime();
return events;
}
if (restartCondition != null && restartCondition.isMeet(nextGen, ndeme)) {
frequencys.setIndexMostPastward(); // start at the very back.
frequencys.getFrequencys(thisGen);
continue;
}
double[] tmp = thisGen;
thisGen = nextGen; // no need for a copy
nextGen = tmp; // copied over. We hope!
// now what about stopping conditions?
// meh..just add a start for now.
}
assert time == model.getStartTime() : time + "\t" + model.getStartTime();
if (time == 0) events.add(new SelectionStartEvent(model.getStartTime(), thisGen));
// System.out.println("FALLThrough:"+events);
return events;
}
/*
* side effect. Uses class variables rather than "passing them" bit "doggy" but saves a sill
* number of arguments.
*/
private void selectionSimulationStep(
double[] previous, double[] next, SelectionStrengthModel[] ssm, double time) {
boolean fixFlag = true;
boolean zeroFlag = true;
// lots of class v set from the calling function.
for (int d = 0; d < ndeme; d++) {
if (sizes[d] == PopulationSizeModel.NULL_POPULATION) continue;
double x = previous[d];
// assumption
double ax = 1.0 - x;
if (x > 0) zeroFlag = false;
if (x < 1.0) fixFlag = false;
double sAA = ssm[d].getStrength(1, 1, time);
double sAa = ssm[d].getStrength(1, 0, time);
double saa = ssm[d].getStrength(0, 0, time);
// System.err.println("SAA @ time:"+time+"\t"+sAA+"\t"+sAa+"\t"+saa);
// System.err.println("x's:"+x+"\t"+ax);
afterSelectionA[d] = x * (1 + ax * sAa + x * sAA);
afterSelectiona[d] = ax * (1 + x * sAa + ax * saa);
}
for (int d = 0; d < ndeme; d++) {
if (sizes[d] == PopulationSizeModel.NULL_POPULATION) {
next[d] = 0; // frequencys[d][1][nextGen] =0;
// nextFreq[d][0]=0;//frequencys[d][0][nextGen] =0;
continue;
}
// first self migration.
double rate = 1 - totalMRates[d];
double nA = rate * afterSelectionA[d];
double na = rate * afterSelectiona[d];
// for(int[] dier:migrationDirections)
// System.out.println(Arrays.toString(dier));
int[] directions = migrationDirections[d];
double[] rates = migrationRates[d];
for (int count = 0; count < directions.length; count++) {
int j = directions[count];
rate = rates[count];
// System.err.println("After!"+Arrays.toString(afterSelectionA)+"\t"+j+"\tD:"+Arrays.toString(directions));
nA += rate * afterSelectionA[j];
na += rate * afterSelectiona[j];
}
// now put it together for xp
double xp = (nA * (1 - nu) + mu * na) / (nA + na);
// we keep N in the sample for now...
int N = (int) Math.ceil(sizes[d].populationSize(time) * 2); //
// System.err.println("CallBin "+N+" "+xp+"\t"+sizes[d].populationSize(time)+"\t"+nA+"\t"+na);
double f = (double) binomial.generateBinomial(N, xp) / N;
// System.err.println("return bin "+f);
// what if N is zero!
if (N == 0) {
f = 0; // throw new RuntimeException("NaN, PopulationSize is probably zero! "+N);
}
// update
next[d] = f;
}
}
private String toString(double[][] freqs) {
String s = "";
for (int i = 0; i < freqs.length; i++) {
s += freqs[i][1] + "\t";
}
return s;
}
private static void initFrequencys(Model parent, SuperFrequencyTrace frequencys) {
// assume frequecys is set to end.
PopulationSizeModel[] sizes = parent.getPopulationSizeModels();
double total = 0;
for (int d = 0; d < sizes.length; d++) {
total += sizes[d].populationSize(0); // assuming homogenous time
// model.
}
double pick = RandomGenerator.getRandom().nextDouble() * total;
int d = 0;
total = 0;
for (int i = 0; i < sizes.length; i++) {
total += sizes[d].populationSize(0);
if (total >= pick) {
d = i;
break;
}
}
int choosen = parent.getModelHistory().getSeedDeme();
if (choosen > -1) d = choosen;
double[] freqs = new double[sizes.length]; // demeCount;
freqs[d] =
1.0 / (parent.getPopulationSizeModels()[d].populationSize(parent.getEndTime())); // FIXME
frequencys.setFrequencys(freqs);
// System.out.println("INITF:"+Arrays.toString(freqs));
}
// public static void initFrequencys(Model model,FrequencyTrace frequencys) {
// double[][] freqs =frequencys.getEnd();
// PopulationSizeModel[] sizes =model.getPopulationSizeModels();
// double total =0;
// for (int d =0; d < freqs.length; d++) {
// freqs[d][1] =0;
// freqs[d][0] =1;
// total +=sizes[d].populationSize(0);// assuming homogenous time model.
//
// }
// double pick =RandomGenerator.getRandom().nextDouble() * total;
// int d =0;
// total =0;
// for (int i =0; i < sizes.length; i++) {
// total +=sizes[d].populationSize(0);
// if (total >= pick) {
// d =i;
// break;
// }
// }
// freqs[d][1] =1.0 /
// (model.getPopulationSizeModels()[d].populationSize(frequencys.getEndTime()) * 4);
// freqs[d][0] =1.0 - freqs[d][1];
//
// }
}
