[beast-mcmc] r6741 committed - Ignore potential infinity problems when using independence samplers.

[beast...@googlecode.com]

Revision: 6741
Author: bael...@gmail.com
Date: Thu Apr 16 14:46:52 2015 UTC
Log: Ignore potential infinity problems when using independence
samplers.
https://code.google.com/p/beast-mcmc/source/detail?r=6741

Modified:
/trunk/src/dr/inference/markovchain/MarkovChain.java

=======================================
--- /trunk/src/dr/inference/markovchain/MarkovChain.java Wed Dec 3
00:44:25 2014 UTC
+++ /trunk/src/dr/inference/markovchain/MarkovChain.java Thu Apr 16
14:46:52 2015 UTC
@@ -25,6 +25,7 @@

package dr.inference.markovchain;

+import dr.evomodel.continuous.GibbsIndependentCoalescentOperator;
import dr.inference.model.CompoundLikelihood;
import dr.inference.model.Likelihood;
import dr.inference.model.Model;
@@ -252,7 +253,9 @@
}

if (score == Double.NEGATIVE_INFINITY && mcmcOperator
instanceof GibbsOperator) {
- Logger.getLogger("error").severe("State " +
currentState + ": A Gibbs opertor, " + mcmcOperator.getOperatorName() + ",
returned a state with zero likelihood.");
+ if (!(mcmcOperator instanceof
GibbsIndependentNormalDistributionOperator) && !(mcmcOperator instanceof
GibbsIndependentGammaOperator) && !(mcmcOperator instanceof
GibbsIndependentCoalescentOperator) && !(mcmcOperator instanceof
GibbsIndependentJointNormalGammaOperator)) {
+ Logger.getLogger("error").severe("State " +
currentState + ": A Gibbs opertor, " + mcmcOperator.getOperatorName() + ",
returned a state with zero likelihood.");
+ }
}

if (score == Double.POSITIVE_INFINITY ||