[beast-mcmc] r4807 committed - Added caching - 10 fold speed increase

[beast...@googlecode.com]

Revision: 4807
Author: rambaut
Date: Fri Mar 9 06:54:57 2012
Log: Added caching - 10 fold speed increase
http://code.google.com/p/beast-mcmc/source/detail?r=4807

Modified:
/trunk/src/dr/evomodel/antigenic/AntigenicLikelihood.java
/trunk/src/dr/inference/model/MatrixParameter.java

=======================================
--- /trunk/src/dr/evomodel/antigenic/AntigenicLikelihood.java Fri Mar 9
03:30:01 2012
+++ /trunk/src/dr/evomodel/antigenic/AntigenicLikelihood.java Fri Mar 9
06:54:57 2012
@@ -182,7 +182,8 @@
if (columnEffectsParameter != null) {
columnEffectsParameter.setDimension(columnLabels.size());
addVariable(columnEffectsParameter);
- String[] labelArray = (String[])columnLabels.toArray();
+ String[] labelArray = new String[columnLabels.size()];
+ columnLabels.toArray(labelArray);

((Parameter.Abstract)columnEffectsParameter).setDimensionNames(labelArray);
}

@@ -190,7 +191,8 @@
if (rowEffectsParameter != null) {
rowEffectsParameter.setDimension(rowLabels.size());
addVariable(rowEffectsParameter);
- String[] labelArray = (String[])rowLabels.toArray();
+ String[] labelArray = new String[rowLabels.size()];
+ rowLabels.toArray(labelArray);

((Parameter.Abstract)rowEffectsParameter).setDimensionNames(labelArray);
}

@@ -217,6 +219,9 @@
}
}

+ locationChanged = new
boolean[this.locationsParameter.getRowDimension()];
+ logLikelihoods = new double[measurements.size()];
+ storedLogLikelihoods = new double[measurements.size()];
likelihoodKnown = false;
}

@@ -236,6 +241,7 @@
@Override
protected void handleVariableChangedEvent(Variable variable, int
index, Variable.ChangeType type) {
if (variable == locationsParameter) {
+ locationChanged[index / mdsDimension] = true;
} else if (variable == mdsPrecisionParameter) {
} else if (variable == columnEffectsParameter) {
} else if (variable == rowEffectsParameter) {
@@ -248,10 +254,15 @@

@Override
protected void storeState() {
+ System.arraycopy(logLikelihoods, 0, storedLogLikelihoods, 0,
logLikelihoods.length);
}

@Override
protected void restoreState() {
+ double[] tmp = logLikelihoods;
+ logLikelihoods = storedLogLikelihoods;
+ storedLogLikelihoods = tmp;
+
likelihoodKnown = false;
}

@@ -280,38 +291,52 @@
double sd = 1.0 / Math.sqrt(precision);

logLikelihood = 0.0;
+ int i = 0;
for (Measurement measurement : measurements) {
- double distance = computeDistance(measurement.rowStrain,
measurement.columnStrain);
-
- double logNormalization =
calculateTruncationNormalization(distance, sd);
-
- switch (measurement.type) {
- case INTERVAL: {
- double minTitre = transformTitre(measurement.minTitre,
measurement.column, measurement.row, distance, sd);
- double maxTitre = transformTitre(measurement.maxTitre,
measurement.column, measurement.row, distance, sd);
- logLikelihood +=
computeMeasurementIntervalLikelihood(minTitre, maxTitre) - logNormalization;
- } break;
- case POINT: {
- double titre = transformTitre(measurement.minTitre,
measurement.column, measurement.row, distance, sd);
- logLikelihood += computeMeasurementLikelihood(titre) -
logNormalization;
- } break;
- case LOWER_BOUND: {
- double minTitre = transformTitre(measurement.minTitre,
measurement.column, measurement.row, distance, sd);
- logLikelihood +=
computeMeasurementLowerBoundLikelihood(minTitre) - logNormalization;
- } break;
- case UPPER_BOUND: {
- double maxTitre = transformTitre(measurement.maxTitre,
measurement.column, measurement.row, distance, sd);
- logLikelihood +=
computeMeasurementUpperBoundLikelihood(maxTitre) - logNormalization;
- } break;
- case MISSING:
- break;
- }
+
+ if (locationChanged[measurement.rowStrain] ||
locationChanged[measurement.columnStrain]) {
+ double distance = computeDistance(measurement.rowStrain,
measurement.columnStrain);
+
+ double logNormalization =
calculateTruncationNormalization(distance, sd);
+
+ switch (measurement.type) {
+ case INTERVAL: {
+ double minTitre =
transformTitre(measurement.minTitre, measurement.column, measurement.row,
distance, sd);
+ double maxTitre =
transformTitre(measurement.maxTitre, measurement.column, measurement.row,
distance, sd);
+ logLikelihoods[i] =
computeMeasurementIntervalLikelihood(minTitre, maxTitre) - logNormalization;
+ } break;
+ case POINT: {
+ double titre =
transformTitre(measurement.minTitre, measurement.column, measurement.row,
distance, sd);
+ logLikelihoods[i] =
computeMeasurementLikelihood(titre) - logNormalization;
+ } break;
+ case LOWER_BOUND: {
+ double minTitre =
transformTitre(measurement.minTitre, measurement.column, measurement.row,
distance, sd);
+ logLikelihoods[i] =
computeMeasurementLowerBoundLikelihood(minTitre) - logNormalization;
+ } break;
+ case UPPER_BOUND: {
+ double maxTitre =
transformTitre(measurement.maxTitre, measurement.column, measurement.row,
distance, sd);
+ logLikelihoods[i] =
computeMeasurementUpperBoundLikelihood(maxTitre) - logNormalization;
+ } break;
+ case MISSING:
+ break;
+ }
+ }
+ logLikelihood += logLikelihoods[i];
+ i++;
}

likelihoodKnown = true;

+ clearLocationChangedFlags();
+
return logLikelihood;
}
+
+ private void clearLocationChangedFlags() {
+ for (int i = 0; i < locationChanged.length; i++) {
+ locationChanged[i] = false;
+ }
+ }

protected double computeDistance(int rowStrain, int columnStrain) {
if (rowStrain == columnStrain) {
@@ -437,6 +462,10 @@
private double logLikelihood = 0.0;
private boolean likelihoodKnown = false;

+ private final boolean[] locationChanged;
+ private double[] logLikelihoods;
+ private double[] storedLogLikelihoods;
+
// **************************************************************
// XMLObjectParser
// **************************************************************
@@ -482,7 +511,10 @@

MatrixParameter locationsParameter = (MatrixParameter)
xo.getElementFirstChild(LOCATIONS);

- Parameter datesParameter = (Parameter)
xo.getElementFirstChild(DATES);
+ Parameter datesParameter = null;
+ if (xo.hasChildNamed(STRAINS)) {
+ datesParameter = (Parameter)
xo.getElementFirstChild(DATES);
+ }

Parameter mdsPrecision = (Parameter)
xo.getElementFirstChild(MDS_PRECISION);

=======================================
--- /trunk/src/dr/inference/model/MatrixParameter.java Tue Sep 27 11:40:39
2011
+++ /trunk/src/dr/inference/model/MatrixParameter.java Fri Mar 9 06:54:57
2012
@@ -68,12 +68,12 @@
}

public int getColumnDimension() {
- return getParameterCount();
+ return getParameter(0).getDimension();
}

public int getRowDimension() {
- return getParameter(0).getDimension();
- }
+ return getParameterCount();
+ }

public String toSymmetricString() {
StringBuffer sb = new StringBuffer("{");