BEAST 1.7.3 is released

[Walter]

Dear Users,

BEAST 1.7.3 is released. 

Thank you for your support,

Walter

[Jacob Berv]

It seems the generate BEAST file button in BEAUTi 1.7.3 is broken somehow. Just set up a new analysis from scratch, and it seems that if I define any taxa to be monophyletic the Generate button does nothing. 

[Jacob Berv]

Tried with 1.7.0 and works fine. I'm on a 2006 MacBook Pro running snow leopard. 

[Andrew Rambaut]

Dear Jacob,

I am sorry but I cannot replicate this problem. I loaded a file, created some taxon sets and set them to be monophyletic. The XML generated fine so the 'generate BEAST file' button isn't generally broken. Please provide a more details of what you did, what your system is etc.? If you can replicate the issue with a test data set or post details of exactly what you need to do to get the error. Ideally posting it to our issue database saves time:

http://code.google.com/p/beast-mcmc/issues/list

Andrew

On 20 Aug 2012, at 20:30, Jacob Berv <jacob...@gmail.com> wrote:

It seems the generate BEAST file button in BEAUTi 1.7.3 is broken somehow. Just set up a new analysis from scratch, and it seems that if I define any taxa to be monophyletic the Generate button does nothing. 

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To view this discussion on the web visit https://groups.google.com/d/msg/beast-users/-/6RwdexxtUpsJ.
To post to this group, send email to beast...@googlegroups.com.
To unsubscribe from this group, send email to beast-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/beast-users?hl=en.

___________________________________________________________________
 Andrew Rambaut                
 Institute of Evolutionary Biology       University of Edinburgh
 Ashworth Laboratories                         Edinburgh EH9 3JT
 EMAIL - a.ra...@ed.ac.uk                TEL - +44 131 6508624  

[Jacob Berv]

I think I've narrowed it down a bit - In order to determine what step was causing the problem I tried clicking the generate button after every step.

I was trying to set up an phylogeographic analysis using the continuous diffusion model.

1. Imported my data (4 genes), 1 start tree - OK 

2. unlinked clocks, models, linked trees - OK 

3. Defined ingroup to be monophyletic - OK 

4. defined lat, long traits for continuous phylogeographic analysis, created partition - OK 

5. set models, clocks, tree prior using logistic growth - OK 

6. set priors - OK

7. set prior for treemodel.rootheight - FAIL

8. remove trait partition - OK

I can replicate this problem if I use the RacRABV dataset (and choose an arbitrary ingroup) so unless I'm missing something this does seem to be a bug with how the traits are managed/defined.

What do you think is going on?

[Jacob Berv]

I should add that I have the same problem for discrete traits as well. 

[Andrew Rambaut]

Thanks Jacob,

What is the prior you set on rootheight?

Andrew

--
You received this message because you are subscribed to the Google Groups "beast-users" group.

To view this discussion on the web visit https://groups.google.com/d/msg/beast-users/-/ht4ekyspQ8gJ.

To post to this group, send email to beast...@googlegroups.com.
To unsubscribe from this group, send email to beast-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/beast-users?hl=en.

[Jacob Berv]

It seems to fail on arbitrary rootheight priors, but I've been using 4.52, 0.851

[Nick Matzke]

Hi all,

I am attempting to sample ancestral states for a sequence of
morphological characters (binary) evolving under a lewisMk
model, with some unknown states.

I can run the analysis just fine until I add the ancestral
state reconstruction code.

The key bits of XML here are:

[...defining the character evolution model...]

<lewisMk id="lewis_un_2_Model">
<frequencies>
<FrequencyModel idref="morph_2_Model"/>
</frequencies>
</lewisMk>

[...define the likelihood of that data on the tree...]

<ancestralTreeLikelihood id="morph.un2.treeLikelihood"
useAmbiguities="true" stateTagName="morph.un2.states">
<patterns idref="morph_un2.patterns"/>
<treeModel idref="treeModel"/>
<siteModel idref="morph_un2.siteModel"/>
<discretizedBranchRates idref="branchRates"/>
<lewisMk idref="lewis_un_2_Model"/>
</ancestralTreeLikelihood>


[...log the ancestral states...]


<!-- START Ancestral state reconstruction -->
<trait name="morph.un2.states" tag="morph.un2.state">
<ancestralTreeLikelihood idref="morph.un2.treeLikelihood"/>
</trait>
<!-- END Ancestral state reconstruction -->








Problems that arise:

1. When I run this on BEAST 1.7.3 WITHOUT Beagle, I get:

======================
[...]

TreeLikelihood(treeModel) using Java general likelihood core
Using ambiguities in tree likelihood.
With 223 unique site patterns.
Branch rate model used: discretizedBranchRates
Ancestral reconstruction using ambiguities is currently not
support without BEAGLE

[crash]
======================







2. But when I run WITH Beagle turned on, I get:

=============================
[...]

Using discretized relaxed clock model.
over sampling = 1
parametric model = logNormalDistributionModel
rate categories = 70

Creating state frequenciesParsing error - poorly formed
BEAST file,
v13a_BDserial_onlyGreatApes_noDNA_add_tip_sampling4.xml:

The '<lewisMk>' element with id, 'lewis_un_2_Model', is
incorrectly constructed.

The following was expected:
Exactly one ELEMENT of name frequencies REQUIRED containing
Exactly one ELEMENT of type FrequencyModel REQUIRED
model 'null': Initial frequencies = {0.5, 0.5}
Creating state frequencies model 'null': Initial frequencies
= {0.33333, 0.33333, 0.33333}
Creating state frequencies model 'null': Initial frequencies
= {0.25, 0.25, 0.25, 0.25}
Creating state frequencies model 'null': Initial frequencies
= {0.2, 0.2, 0.2, 0.2, 0.2}

BEAST has terminated with an error. Please select QUIT from
the menu.
=============================

The second error occurs despite the fact that the lewisMk
model is specified just as required (see above) and the
identical XML works fine in non-state-reconstruction analyses.

It looks like the problem is also noted on CIPRES:

http://www.phylo.org/index.php/help/error_messages/#beastbeagle






3. Just for kicks I also tried running ancestral state
reconstruction with beagle off, AND changing to
"useAmbiguities=false". This gave a hard crash:

=============================
Branch rate model used: discretizedBranchRates
Creating swap operator for parameter branchRates.categories
(weight=10.0)
Likelihood computation is using an auto sizing thread pool.
Creating the MCMC chain:
chainLength=100000000
autoOptimize=true
autoOptimize delayed for 1000000 steps
# BEAST v1.7.3, rr5146
# Generated Mon Aug 20 19:25:37 PDT 2012 [seed=1345515937166]
state Posterior Prior Likelihood Root Height
Rate
0 -3316.5324 -313.3966 -3003.1358 32.95
1.00000 -
Exception in thread "Thread-2" java.lang.RuntimeException:
Reconstruction not implemented for multiple categories yet.

at
dr.evomodel.treelikelihood.AncestralStateTreeLikelihood.traverseSample(Unknown
Source)
at
dr.evomodel.treelikelihood.AncestralStateTreeLikelihood.traverseSample(Unknown
Source)
at
dr.evomodel.treelikelihood.AncestralStateTreeLikelihood.redrawAncestralStates(Unknown
Source)
at
dr.evomodel.treelikelihood.AncestralStateTreeLikelihood.getStatesForNode(Unknown
Source)
at
dr.evomodel.treelikelihood.AncestralStateTreeLikelihood$1.getTraitString(Unknown
Source)
at
dr.evomodelxml.tree.TreeLoggerParser$2.getTraitString(Unknown Source)
at dr.evolution.tree.Tree$Utils.writeTreeTraits(Unknown Source)
at dr.evolution.tree.Tree$Utils.newick(Unknown Source)
at dr.evolution.tree.Tree$Utils.newick(Unknown Source)
at dr.evolution.tree.Tree$Utils.newick(Unknown Source)
at dr.evolution.tree.Tree$Utils.newick(Unknown Source)
at dr.evolution.tree.Tree$Utils.newick(Unknown Source)
at dr.evomodel.tree.TreeLogger.log(Unknown Source)
at dr.inference.mcmc.MCMC$1.currentState(Unknown Source)
at
dr.inference.markovchain.MarkovChain.fireCurrentModel(Unknown Source)
at dr.inference.markovchain.MarkovChain.runChain(Unknown Source)
at dr.inference.mcmc.MCMC.chain(Unknown Source)
at dr.inference.mcmc.MCMC.run(Unknown Source)
at java.lang.Thread.run(Thread.java:680)
=============================

...which may be due to having ambiguities in the data but
saying "useAmbiguities=false" anway.

The above problems may just be "stuff that hasn't been
implemented yet", but I figured I should summarize my
findings to help the google-ability of this issue for
others, and/or to find out/hope that the wonderful BEAST
developers have something in the works to cover this.

(Or maybe I missed something obvious.)

Cheers!
Nick

[Jacob Berv]

The workaround seems to be to produce an XML file without setting the rootheight prior, an XML file with the rootheight prior but without the phylogeography setup, and then to copy the priors block from the latter over the former. 

[Andrew Rambaut]

Thanks Jacob,

I will investigate what the issue is.

Andrew

To view this discussion on the web visit https://groups.google.com/d/msg/beast-users/-/l9QQyV5ffwgJ.

To post to this group, send email to beast...@googlegroups.com.
To unsubscribe from this group, send email to beast-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/beast-users?hl=en.

[Jacob Berv]

So while my workaround seems to have gotten the XML to run.. after 24 hours it seems something really strange is happening during the analysis. After a few million generations the rootheight gets fixed on a single value and continues for the remainder of the run. This gives a straight line in the trace file. The populations I'm analyzing have multiple individuals per sampling site so I added the jitter code with a window of 0.0005 to add some noise to the localities in the multivariatetraitlikelihood block as follows:

<jitter window="0.0005 0.0005" duplicatesOnly="true">

<parameter idref="leaf.location"/>

</jitter>

Any idea what could be causing this? 

[Jacob Berv]

Hi Andrew - have you made any headway on this issue in the last 10 days?

[Andrew Rambaut]

What issue?

Andrew

On 31 Aug 2012, at 00:45, Jacob Berv <jacob...@gmail.com> wrote:

> Hi Andrew - have you made any headway on this issue in the last 10 days?
>

> --
> You received this message because you are subscribed to the Google Groups "beast-users" group.

> To view this discussion on the web visit https://groups.google.com/d/msg/beast-users/-/vkXVJDJbrhgJ.

> To post to this group, send email to beast...@googlegroups.com.
> To unsubscribe from this group, send email to beast-users...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/beast-users?hl=en.

___________________________________________________________________
Andrew Rambaut
Institute of Evolutionary Biology University of Edinburgh
Ashworth Laboratories Edinburgh EH9 3JT
EMAIL - a.ra...@ed.ac.uk TEL - +44 131 6508624





--

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

[Jacob Berv]

Well at first it was the issue with the generate beast xml button that I described earlier in the thread. I haven't found a way around this bug beyond producing two xml files and combining them. A few days later, after successfully getting the combined files to actually run, I had an issue with strange artifacts occurring the run. For example, the treeRoot would get stuck on a single value and not update between generations. Setting beagle scaling equal to always seems to have solved this issue. Now, I'm gotten my files to run and go to completion, but I'm having serious convergence problems with the phylogeographic analysis and some other strange issues.

I have tried discrete analysis (areas of endemism) and continuous analysis (sampling coordinates) (brownian, cauchy RRW and gamma RRW, all produced in 1.7.3). The dataset I've been working with normally is well converged by 100M generations, sampled every 10k. With the addition of the discrete analysis, it seems I need to bump it to 200M generations (sampled every 10k) to get convergence and ESS's above 200 for all parameters. For the continuous analysis (brownian), the ESS for the prior and likelihood (and others) are mostly under 30, even after 200M generations. I checked it at 100M and 200M generations and the ESSs were mostly unchanged. For the cauchy and gamma RRW analysis, I get high ESS values and convergence by 200M generations, but only the cauchy RRW makes any sense when I visualize it in spread. The gamma RRW suggests several ancestral locations that are totally impossible (and far outside of the range of the original coordinates), and dispersal routes that zig zag around the map. In sum, for the continuous analysis, I have only been able to get the cauchy RRW model to work so far. When I do a bayes factor test in Tracer on the brownian, cauchy, and gamma log files, it suggests the gamma model is best, but as I mentioned, the visualization of ancestral locations is not plausible. Any suggestions would be greatly appreciated. I would be happy to send you my XML files directly if you think that would help.

Finally, another bug I've encountered in 1.7.3 is whenever I do a run, I get a bunch of "Unhandled attribute (typo?) stateTagName in treeLikelihood" messages, though I've been ignoring them and it doesn't seem to make a difference in the analysis... 

[Nelson Naranjo]

HI Jacob,

I want to know how could you solve the problem of "outside of the range of the original coordinates with the gamma RRW model".

Regard

Nelson