use of dr.evomodel.substmodel.MarkovJumpsSubstitutionModel in project beast-mcmc by beast-dev.
the class MarkovJumpsBeagleTreeLikelihood method addRegister.
public void addRegister(Parameter addRegisterParameter, MarkovJumpsType type, boolean scaleByTime) {
if ((type == MarkovJumpsType.COUNTS && addRegisterParameter.getDimension() != stateCount * stateCount) || (type == MarkovJumpsType.REWARDS && addRegisterParameter.getDimension() != stateCount)) {
throw new RuntimeException("Register parameter of wrong dimension");
}
addVariable(addRegisterParameter);
final String tag = addRegisterParameter.getId();
for (int i = 0; i < substitutionModelDelegate.getSubstitutionModelCount(); ++i) {
registerParameter.add(addRegisterParameter);
MarkovJumpsSubstitutionModel mjModel;
SubstitutionModel substitutionModel = substitutionModelDelegate.getSubstitutionModel(i);
if (useUniformization) {
mjModel = new UniformizedSubstitutionModel(substitutionModel, type, nSimulants);
} else {
if (type == MarkovJumpsType.HISTORY) {
throw new RuntimeException("Can only report complete history using uniformization");
}
mjModel = new MarkovJumpsSubstitutionModel(substitutionModel, type);
}
markovjumps.add(mjModel);
branchModelNumber.add(i);
addModel(mjModel);
setupRegistration(numRegisters);
String traitName;
if (substitutionModelDelegate.getSubstitutionModelCount() == 1) {
traitName = tag;
} else {
traitName = tag + i;
}
jumpTag.add(traitName);
expectedJumps.add(new double[treeModel.getNodeCount()][patternCount]);
// storedExpectedJumps.add(new double[treeModel.getNodeCount()][patternCount]);
boolean[] oldScaleByTime = this.scaleByTime;
int oldScaleByTimeLength = (oldScaleByTime == null ? 0 : oldScaleByTime.length);
this.scaleByTime = new boolean[oldScaleByTimeLength + 1];
if (oldScaleByTimeLength > 0) {
System.arraycopy(oldScaleByTime, 0, this.scaleByTime, 0, oldScaleByTimeLength);
}
this.scaleByTime[oldScaleByTimeLength] = scaleByTime;
if (type != MarkovJumpsType.HISTORY) {
TreeTrait.DA da = new TreeTrait.DA() {
final int registerNumber = numRegisters;
public String getTraitName() {
return tag;
}
public Intent getIntent() {
return Intent.BRANCH;
}
public double[] getTrait(Tree tree, NodeRef node) {
return getMarkovJumpsForNodeAndRegister(tree, node, registerNumber);
}
};
treeTraits.addTrait(traitName + "_base", da);
treeTraits.addTrait(addRegisterParameter.getId(), new TreeTrait.SumAcrossArrayD(new TreeTrait.SumOverTreeDA(da)));
} else {
if (histories == null) {
histories = new String[treeModel.getNodeCount()][patternCount];
} else {
throw new RuntimeException("Only one complete history per markovJumpTreeLikelihood is allowed");
}
if (nSimulants > 1) {
throw new RuntimeException("Only one simulant allowed when saving complete history");
}
// Add total number of changes over tree trait
TreeTrait da = new TreeTrait.DA() {
final int registerNumber = numRegisters;
public String getTraitName() {
return tag;
}
public Intent getIntent() {
return Intent.BRANCH;
}
public double[] getTrait(Tree tree, NodeRef node) {
return getMarkovJumpsForNodeAndRegister(tree, node, registerNumber);
}
};
treeTraits.addTrait(addRegisterParameter.getId(), new TreeTrait.SumOverTreeDA(da));
// Record the complete history for this register
historyRegisterNumber = numRegisters;
((UniformizedSubstitutionModel) mjModel).setSaveCompleteHistory(true);
if (useCompactHistory && logHistory) {
treeTraits.addTrait(ALL_HISTORY, new TreeTrait.SA() {
public String getTraitName() {
return ALL_HISTORY;
}
public Intent getIntent() {
return Intent.BRANCH;
}
public boolean getFormatAsArray() {
return true;
}
public String[] getTrait(Tree tree, NodeRef node) {
List<String> events = new ArrayList<String>();
for (int i = 0; i < patternCount; i++) {
String eventString = getHistoryForNode(tree, node, i);
if (eventString != null && eventString.compareTo("{}") != 0) {
eventString = eventString.substring(1, eventString.length() - 1);
if (eventString.contains("},{")) {
// There are multiple events
String[] elements = eventString.split("(?<=\\}),(?=\\{)");
for (String e : elements) {
events.add(e);
}
} else {
events.add(eventString);
}
}
}
String[] array = new String[events.size()];
events.toArray(array);
return array;
}
public boolean getLoggable() {
return true;
}
});
}
for (int site = 0; site < patternCount; ++site) {
final String anonName = (patternCount == 1) ? HISTORY : HISTORY + "_" + (site + 1);
final int anonSite = site;
treeTraits.addTrait(anonName, new TreeTrait.S() {
public String getTraitName() {
return anonName;
}
public Intent getIntent() {
return Intent.BRANCH;
}
public String getTrait(Tree tree, NodeRef node) {
String history = getHistoryForNode(tree, node, anonSite);
// Return null if empty
return (history.compareTo("{}") != 0) ? history : null;
}
public boolean getLoggable() {
return logHistory && !useCompactHistory;
}
});
}
}
numRegisters++;
}
// End of loop over branch models
}
use of dr.evomodel.substmodel.MarkovJumpsSubstitutionModel in project beast-mcmc by beast-dev.
the class MarkovJumpsBeagleTreeLikelihood method hookCalculation.
protected void hookCalculation(Tree tree, NodeRef parentNode, NodeRef childNode, int[] parentStates, int[] childStates, double[] inProbabilities, int[] rateCategory) {
final int childNum = childNode.getNumber();
double[] probabilities = inProbabilities;
if (probabilities == null) {
// Leaf will call this hook with a null
getMatrix(childNum, tmpProbabilities);
probabilities = tmpProbabilities;
}
final double branchRate = branchRateModel.getBranchRate(tree, childNode);
final double parentTime = tree.getNodeHeight(parentNode);
final double childTime = tree.getNodeHeight(childNode);
final double substTime = parentTime - childTime;
for (int r = 0; r < markovjumps.size(); r++) {
MarkovJumpsSubstitutionModel thisMarkovJumps = markovjumps.get(r);
final int modelNumberFromrRegistry = branchModelNumber.get(r);
// int dummy = 0;
// final int modelNumberFromTree = branchSubstitutionModel.getBranchIndex(tree, childNode, dummy);
// @todo AR - not sure about this - if this is an epoch this is just going to get the most
// @todo tipward model for the branch. I think this was what was happening before (in comment,
// @todo above).
BranchModel.Mapping mapping = branchModel.getBranchModelMapping(childNode);
if (modelNumberFromrRegistry == mapping.getOrder()[0]) {
if (useUniformization) {
computeSampledMarkovJumpsForBranch(((UniformizedSubstitutionModel) thisMarkovJumps), substTime, branchRate, childNum, parentStates, childStates, parentTime, childTime, probabilities, scaleByTime[r], expectedJumps.get(r), rateCategory, r == historyRegisterNumber);
} else {
computeIntegratedMarkovJumpsForBranch(thisMarkovJumps, substTime, branchRate, childNum, parentStates, childStates, probabilities, condJumps, scaleByTime[r], expectedJumps.get(r), rateCategory);
}
} else {
// Fill with zeros
double[] result = expectedJumps.get(r)[childNum];
Arrays.fill(result, 0.0);
}
}
}
use of dr.evomodel.substmodel.MarkovJumpsSubstitutionModel in project beast-mcmc by beast-dev.
the class UniformizedStateHistoryTest method testStateHistorySimulationForJumps.
public void testStateHistorySimulationForJumps() {
try {
double startingTime = 1.0;
double endingTime = 3.0;
int startingState = 1;
int endingState = 3;
int N = 1000000;
double[] tmp = new double[stateCount * stateCount];
hky.getTransitionProbabilities(endingTime - startingTime, tmp);
double transitionProbability = tmp[startingState * stateCount + endingState];
double[][] registers = new double[2][stateCount * stateCount];
// Count all jumps
MarkovJumpsCore.fillRegistrationMatrix(registers[0], stateCount);
// Mark just one state!
registers[1][2 * stateCount + 1] = 1.0;
double[] expectations = new double[registers.length];
for (int i = 0; i < N; i++) {
StateHistory history = UniformizedStateHistory.simulateConditionalOnEndingState(startingTime, startingState, endingTime, endingState, transitionProbability, stateCount, process);
for (int j = 0; j < registers.length; j++) {
expectations[j] += history.getTotalRegisteredCounts(registers[j]);
}
}
// Determine analytic solution
MarkovJumpsSubstitutionModel markovjumps = new MarkovJumpsSubstitutionModel(hky);
double[] mjExpectations = new double[stateCount * stateCount];
for (int j = 0; j < registers.length; j++) {
expectations[j] /= (double) N;
System.out.println("Expected number for register = " + expectations[j]);
markovjumps.setRegistration(registers[j]);
markovjumps.computeCondStatMarkovJumps(endingTime - startingTime, mjExpectations);
assertEquals(mjExpectations[startingState * stateCount + endingState], expectations[j], 1E-2);
}
} catch (SubordinatedProcess.Exception e) {
throw new RuntimeException("Subordinated process exception");
}
}
use of dr.evomodel.substmodel.MarkovJumpsSubstitutionModel in project beast-mcmc by beast-dev.
the class StateHistoryTest method setUp.
public void setUp() {
MathUtils.setSeed(666);
freqModel = new FrequencyModel(Nucleotides.INSTANCE, new double[] { 0.45, 0.25, 0.05, 0.25 });
baseModel = new HKY(2.0, freqModel);
stateCount = baseModel.getDataType().getStateCount();
lambda = new double[stateCount * stateCount];
baseModel.getInfinitesimalMatrix(lambda);
System.out.println("lambda = " + new Vector(lambda));
markovjumps = new MarkovJumpsSubstitutionModel(baseModel);
}
use of dr.evomodel.substmodel.MarkovJumpsSubstitutionModel in project beast-mcmc by beast-dev.
the class MarkovJumpsSubstitutionModelTest method testMarginalRates.
public void testMarginalRates() {
HKY substModel = new HKY(2.0, new FrequencyModel(Nucleotides.INSTANCE, // A,C,G,T
new double[] { 0.3, 0.2, 0.25, 0.25 }));
int states = substModel.getDataType().getStateCount();
MarkovJumpsSubstitutionModel markovjumps = new MarkovJumpsSubstitutionModel(substModel, MarkovJumpsType.COUNTS);
double[] r = new double[states * states];
// A
int from = 0;
// C
int to = 1;
MarkovJumpsCore.fillRegistrationMatrix(r, from, to, states, 1.0);
markovjumps.setRegistration(r);
double marginalRate = markovjumps.getMarginalRate();
System.out.println("Marginal rate = " + marginalRate);
assertEquals(rMarkovMarginalRate, marginalRate, tolerance);
MarkovJumpsCore.fillRegistrationMatrix(r, states);
markovjumps.setRegistration(r);
marginalRate = markovjumps.getMarginalRate();
System.out.println("Marginal rate = " + marginalRate);
assertEquals(1.0, marginalRate, tolerance);
}
Aggregations