CIPRES run problems using BEAGLE and BEAST - terminates after two cycles.
Mark Schultz
each time, even though the same .xml worked perfectly on my local
machine. CIPRES were very helpful, but in the end they asked me if I
could solve my problem via the BEAST forum and get back to them.
Here's what they wrote:
"We tried a fix suggested in the beast user group, but no success.
Part of the problem may be relatively few patterns for the data.
But I am honestly not sure.
In the BEAST user page, this advice is offered in response to the error message:
"if you are brave, you could ignore these very small (and
hopefully insignificant) differences by setting the <mcmc> XML block
attribute fullEvaluation="0". "
I cannot verify that this is good advice, but you might try contacting
the beast user group for help.
It is clear it is a problem with BEAGLE, but we will have to get help
from those developers to understand/address this issue.
Sorry I dont have more concrete information,"
They went on to write:
"I just want to be clear, this is not a problem we can fix at CIPRES,
it is an issue with Beagle. If you report it to the beast users group,
they may fix it, but I wont report an issue involving your data set
to them, because I dont share user data with anyone."
And then:
"Most BEAST data sets are working fine on CIPRES, but there is
something peculiar about yours.
On the BEAST discussion group, they refer to the error message you
saw, as I mentioned. The developers offered a couple of fixes, and
explained why it happens (see the link I sent you below). One fix is
to modify your input request, as I mentioned, but there is no
guarantee it will work. I may try that, but you can also.
The second was to try and run BEAGLE in a slightly different way,
which failed, as we saw.
Hence, I advised that you show your data set to the beast google user
group to ask them for advice.
The thing is, each of these programs have occasional bugs with certain
data sets. Unfortunately you drew the short straw on this data set.
But although I am pretty good at keeping the portal running, and I
know a few tricks, the people who write these codes are always going
to be better at explaining why your data set is problematic. Moreover,
these codes are always experimental. If you present them with a
failure case, the odds are good they know how to run it correctly, or
you help them make their code better (for the whole community) by
fixing the bug.
It is true that I could take your data set to them and ask what is wrong,
but that violates your confidentiality. This is your data. Also, by
taking the problem to them yourself, you will learn more about the
software.
Again, I encourage you to inquire to the beast user group about this issue
if you want to follow up on it. I will be happy to update our code in
response to any bug fixes that are required....."
On Wed, Oct 19, 2011 at 4:27 AM, Tod W Reeder
<tre...@sunstroke.sdsu.edu> wrote:
> Hi Mark,
> Did anyone chime in with a possible reason for that error message?
> I've been getting the exact error message when trying to do a *BEAST
> analysis on Cipres. In my case, I used BEAUTi 1.6.2 to make my input file.
> Plus, my file includes 42 protein coding loci, each partitioned by codon;
> thus a total of 126 partitions. The input file does execute and run OK on
> my desktop version of *BEAST 1.6.2
> Tod
>
>
> ---------------------------------------------------
> Tod W. Reeder
> Associate Chair & Graduate Advisor
> Department of Biology
> San Diego State University
> San Diego, CA 92182-4614
> 619-594-7826 or 619-594-6375 (office)
> 619-594-5676 (fax)
> tre...@sunstroke.sdsu.edu
> http://sites.google.com/site/todreederslab/
> ---------------------------------------------------
> On Oct 14, 2011, at 8:44 AM, Miller, Mark wrote:
>
> Hi,
>
> We are trying to run a set of files for a user. The files were created using
> BEAUTi 1.6.1. The files work fine on her home machine, and our machine using
> BEAST alone.
> However, under BEAST/BEAGLE we are seeing the error message below. Other
> files work fine using BEAST/BEAGLE.
>
> Fatal exception: dr.evolution.alignment.Patterns cannot be cast to
> dr.evolution.alignment.SitePatterns
> java.lang.ClassCastException: dr.evolution.alignment.Patterns cannot be cast
> to dr.evolution.alignment.SitePatterns
> at
> dr.app.beagle.evomodel.parsers.TreeLikelihoodParser.parseXMLObject(Unknown
> Source)
> at dr.xml.AbstractXMLObjectParser.parseXMLObject(Unknown Source)
> at dr.xml.XMLParser.convert(Unknown Source)
> at dr.xml.XMLParser.convert(Unknown Source)
> at dr.xml.XMLParser.parse(Unknown Source)
> at dr.app.beast.BeastMain.<init>(Unknown Source)
> at dr.app.beast.BeastMain.main(Unknown Source)
>
> Here is the treelikelihood statement in the input files that fail:
>>>
>>> <treeLikelihood id="CP1+2.treeLikelihood" useAmbiguities="false">
>>> <mergePatterns idref="CP1+2.patterns"/>
>>> <treeModel idref="treeModel"/>
>>> <siteModel idref="CP1+2.siteModel"/>
>>> <discretizedBranchRates idref="branchRates"/>
>>> </treeLikelihood>
>>> <treeLikelihood id="CP3.treeLikelihood" useAmbiguities="false">
>>> <mergePatterns idref="CP3.patterns"/>
>>> <treeModel idref="treeModel"/>
>>> <siteModel idref="CP3.siteModel"/>
>>> <discretizedBranchRates idref="branchRates"/>
>>> </treeLikelihood>
>
> I wonder if anyone has an idea about what might be wrong.
>
> Thanks for any advice.
>
> Mark
> --
> You received this message because you are subscribed to the Google Groups
> "beast-users" group.
> 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.
>
> --
> You received this message because you are subscribed to the Google Groups
> "beast-users" group.
> 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
This sounds like an issue with likelihood calculation underflow on BEAGLE. The solution (hopefully) is to add the option '-beagle_scaling always' to the BEAST command line.
Andrew
Miller, Mark
And thanks for the response. The initial run gave this error:
Oct 5, 2011 5:04:39 PM dr.inference.markovchain.MarkovChain runChain
SEVERE: State was not correctly calculated after an operator move.
Likelihood evaluation: -54984.69904267142
Full Likelihood evaluation: -54939.67665352418
Operator: dr.evomodel.operators.ExchangeOperator@d4dd758 Narrow Exchange(treeModel)
We then delivered the command line as follows:
beastwrapper -seed 12345 -beagle_scaling always -beagle_SSE -beagle_instances 8 infile.xml
and got basically the same error message.
Oct 19, 2011 2:08:04 PM dr.inference.markovchain.MarkovChain runChain
SEVERE: State was not correctly calculated after an operator move.
Likelihood evaluation: -55380.07720802788
Full Likelihood evaluation: -55415.890567276445
Operator: uniformOperator(nodeHeights(treeModel)) uniform(nodeHeights(treeModel))
Any idea what we might be doing wrong?
Mark