beast2-dev / beastlabs Goto Github PK

In Slice, the method below is alway returning the 1st element of a parameter, even if the parameter has multiple dimensions:

    public double getArrayValue() {
        return functionInput.get().getArrayValue(0);
    }

The correct code should be

    public double getArrayValue() {
        return functionInput.get().getArrayValue(indexStart);
    }

If a tree contains dated tips the scale method of the Tree can return a tree with negative branch lengths. The operator should check that such tree is not valid and return -Infinity if a negative branch length is detected.

A tree stretch move instead of a scale move might be better regardless.

Need to be confirmed and found out why this happens.

PrunedAlignment uses a given TaxonSet based on whether taxonSet is null or not. However, the corresponding input is optional with a default value of TaxonSet(). This means that PA's initAndValidate should instead check whether taxonSet is empty.

In the main beast2 package there is also test.beast.evolution.likelihood.TreeLikelihoodTest class.

Hello,

I am using the method Particle Filter explained here, and I am facing a deadlock of the threads of the particles when getting into stepsize. Here is the output of "tail particle*/beast.log":

==> particle0/beast.log <==
         Sample      posterior ESS(posterior)          prior     ESS(prior)     likelihood ESS(lik_ihood)
              0    -21975.2366              N       -78.0244              N    -21897.2122              N --
          10000     -4873.4365         2.0         -189.1466         2.0        -4684.2899         2.0    --
          20000     -4480.6041         3.0         -179.8027         3.0        -4300.8014         3.0    --
25000: writing /home/dreyes/prueba_antonio/sideritis_mcmc/particles_cpus/particleset3/particle0/particlelock0
25000: waiting for /home/dreyes/prueba_antonio/sideritis_mcmc/particles_cpus/particleset3/particle0/threadlock0

==> particle1/beast.log <==
         Sample      posterior ESS(posterior)          prior     ESS(prior)     likelihood ESS(lik_ihood)
              0    -21975.1679              N       -78.0244              N    -21897.1434              N --
          10000     -4844.3121         2.0         -216.0260         2.0        -4628.2860         2.0    --
          20000     -4431.8588         3.0         -168.5222         3.0        -4263.3365         3.0    --
25000: writing /home/dreyes/prueba_antonio/sideritis_mcmc/particles_cpus/particleset3/particle1/particlelock0
25000: waiting for /home/dreyes/prueba_antonio/sideritis_mcmc/particles_cpus/particleset3/particle1/threadlock0

==> particle2/beast.log <==
         Sample      posterior ESS(posterior)          prior     ESS(prior)     likelihood ESS(lik_ihood)
              0    -21975.1738              N       -78.0244              N    -21897.1493              N --
          10000     -4690.5879         2.0         -196.0932         2.0        -4494.4947         2.0    --
          20000     -4463.5546         3.0         -162.7462         3.0        -4300.8084         3.0    --
25000: writing /home/dreyes/prueba_antonio/sideritis_mcmc/particles_cpus/particleset3/particle2/particlelock0
25000: waiting for /home/dreyes/prueba_antonio/sideritis_mcmc/particles_cpus/particleset3/particle2/threadlock0

I have read the code and the deadlock probably happens somewhere in between the lines 60 to 69 of ParticleLauncherByFile.java. I have attached my code in a txt because github doesn't allow to attach xml files -> SIDERITIS.MultipleParticles.txt.

Best regards,

Daniel Reyes

Running ant junitBEASTlabs fails DocumentationTest for some BEASTLabs classes.

See below for Junit report details.

Reproducable through running XML from CompEvol/beast2#1055

329e48f fixes the symptoms, but it would be good to prevent short branch lengths from happening during the calculation of the distance in RNNIMetric.

@rbouckaert
PrunedRelaxedClockModel has the code to assume UCRelaxedClockModel meanRate is RealParameter. I add the code to cast it into RealParameter after meanRate is changed to Function .

Please review my bug fix and close this issue if you are happy with the solution.

Binary class files and other stuff have wound up in the repository under /bin. These were introduced as part of commits a6d388d, f0c2a00 and cafaea7.

uncomment the commented code in calcLogP in WeightedDirichlet, and add Unit test to test logP

The starting tree subtemplates provided by BEASTLabs merge with "parametricDistributions" as mergepoint so you cannot remove them from a custom template without also removing the Uniform/etc distributions

Proposed fix: starting tree subtemplates should probably have a specific mergepoint, like the tree priors

In my beast.xml (attached), I have MRCA priors set up in a consistent manner, but due to the implementation of SimpleRandomTree, the initialization fails.

At its core, I have the following stack of MRCA priors.

    <distribution id="alor1247-besar_originateMRCA" monophyletic="true" spec="beast.math.distributions.MRCAPrior" tree="@Tree.t:beastlingTree" useOriginate="true">
      <taxonset id="tx_alor1247-besar" spec="TaxonSet">
        <taxon idref="alor1247-besar" />
      </taxonset>
      <Normal id="DistributionForalor1247-besar_originateMRCA" mean="450.0" name="distr" offset="0.0" sigma="25.5102040816" />
    </distribution>
    <distribution id="alorese_originateMRCA" monophyletic="true" spec="beast.math.distributions.MRCAPrior" tree="@Tree.t:beastlingTree" useOriginate="true">
      <taxonset id="alorese" spec="TaxonSet">
        <plate range="alor1247-baran,alor1247-besar,alor1247-munas,alor1247-pandai" var="language">
          <taxon idref="$(language)" />
        </plate>
      </taxonset>
      <Normal id="DistributionForalorese_originateMRCA" mean="675.0" name="distr" offset="0.0" sigma="12.7551020408" />
    </distribution>
    <distribution id="lamaholoticMRCA" monophyletic="true" spec="beast.math.distributions.MRCAPrior" tree="@Tree.t:beastlingTree">
      <taxonset id="lamaholotic" spec="TaxonSet">
        <plate range="alor1247-baran,alor1247-besar,alor1247-munas,alor1247-pandai,lama1277-adona,lama1277-baipi,lama1277-bama,lama1277-belan,lama1277-botun,lama1277-dulhi,lama1277-horow,lama1277-ileap,lama1277-imulo,lama1277-kalik,lama1277-kiwan,lama1277-lamah,lama1277-lamak,lama1277-lamal,lama1277-lamat,lama1277-lerek,lama1277-lewob,lama1277-lewoe,lama1277-lewog,lama1277-lewoi,lama1277-lewok,lama1277-lewom,lama1277-lewop,lama1277-lewot,lama1277-lewuk,lama1277-merde,lama1277-minga,lama1277-mulan,lama1277-paina,lama1277-pukau,lama1277-ritae,lama1277-tanju,lama1277-waiba,lama1277-waiwa,lama1277-watan,lama1277-wuake" var="language">
          <taxon idref="$(language)" />
        </plate>
      </taxonset>
      <Uniform id="DistributionForlamaholoticMRCA" lower="300.0" name="distr" offset="0.0" upper="9223372036854775807" />
    </distribution>

In the doTheWork() method of SimpleRandomTree, Alor-Besar ist first grouped with one of its Alorese siblings to ensure that the MRCA prior on its ancestor can be fulfilled. Both are then removed from the taxonsets that still need to be considered. When the turn comes op for the equivalent step for the Alorese taxonset, a similar thing happens to another Alorese sibling together with one Lamaholot taxon, but for some reason which still escapes my debugging, that grouping has priority over the Alorese-internal grouping, so the SimpleRandomTree ends up containing something like

(((alor1247-besar,alor1247-baran),(alor1247-munas,(alor1247-pandai,lama1277-lamah))),(lama…

@tgvaughan @gavryushkina
I think this may be interesting to you. I created TreeAnnotation when I tried to colour SA trees by a meta data instead of labels. I also updated Sasha's ConvertZBTreeToSATree to use the meta data mapping file.

TreeAnnotation annotates nodes of a tree according to an annotation map, which is imported from a tab delimited file with header (1 trait only at the moment). The 1st column is tip labels, and the 2nd is trait values.

The annotation map uses tip labels as key, and trait value as value. The trait name is given by the constructor. It can be null, in that case, then the 2nd column name from header is assigned to the trait name.

Usage: TreeAnnotation treeFile mapFilePath

• BernoulliDistribution
• WeightedDirichlet
• BEASTVector : a vector of beast objects

This makes it impossible to use references to BEASTObjects as keys, even when those BEASTObjects represent things like a Taxon or Locus.

Maybe instead of serializing, regexing and deserializing MCMCMC could work entirely with the BEASTObjects themselves. Only CalculationNodes and StateNodes need to be copied after all...

beast.util.LogAnalyserAdv should not :

import test.beast.beast2vs1.trace.LogFileTraces;
import test.beast.beast2vs1.trace.TraceException;
import test.beast.beast2vs1.trace.TraceStatistics;

It should either use beast.util.LogAnalyser or be removed.

add the citation to MCMCMC:

@Citation(value= "Altekar G, Dwarkadas S, Huelsenbeck J and Ronquist F (2004). \n" +
		"  Parallel Metropolis Coupled Markov Chain Monte Carlo For Bayesian Phylogenetic Inference.\n" +
		"  Bioinformatics, 20(3), 407-415."
		, year = 2004, firstAuthorSurname = "Altekar",
		DOI="10.1093/bioinformatics/btg427")

I have written and tested an AncestralStatesLogger extension, which allows the reconstruction at multiple MRCA nodes, each specified by a taxonset, and which logs the original representation of the value instead of the internal one (eg. covarion logs as 01ab instead of 0123 – The code for that was taken from the AncestralStateTreeLikelihood class in beast-classics).

That class should be backward compatible with AncestralStateLogger and I would be happy to create a pull-request to it for this directory, if you like.