Welcome, Guest. Please Login
YaBB - Yet another Bulletin Board
 
  HomeHelpSearchLogin  
 
ModelSelectorCodon.bf bug? (Read 2027 times)
Matt
YaBB Newbies
*
Offline


Feed your monkey!

Posts: 23
ModelSelectorCodon.bf bug?
Aug 9th, 2010 at 5:43pm
 
Using either HYPHYMPI or HYPHYMP....

my bf:
Code:
inputOptions = {};
inputOptions ["00"] = "3";
inputOptions ["01"] = "Universal";
inputOptions ["02"] = "1";
inputOptions ["03"] = DIRECTORY_SEPARATOR + "home" + DIRECTORY_SEPARATOR + "macmanes" + DIRECTORY_SEPARATOR + "Desktop" + DIRECTORY_SEPARATOR + "genus" + DIRECTORY_SEPARATOR + "seqs" + DIRECTORY_SEPARATOR + "Pero.dqa.genus.phy";
inputOptions ["04"] = DIRECTORY_SEPARATOR + "home" + DIRECTORY_SEPARATOR + "macmanes" + DIRECTORY_SEPARATOR + "Desktop" + DIRECTORY_SEPARATOR + "genus" + DIRECTORY_SEPARATOR + "tree" + DIRECTORY_SEPARATOR + "Pero.dqa.genus.newick";
inputOptions ["05"] = "010010";
inputOptions ["06"] = "1";
inputOptions ["07"] = DIRECTORY_SEPARATOR + "home" + DIRECTORY_SEPARATOR + "macmanes" + DIRECTORY_SEPARATOR + "Desktop" + DIRECTORY_SEPARATOR + "genus" + DIRECTORY_SEPARATOR + "hyphy.res" + DIRECTORY_SEPARATOR + "model.selector.codon.out";
inputOptions ["08"] = "c-AIC";
inputOptions ["09"] = "Rank";
inputOptions ["10"] = "Completely Random";
inputOptions ["11"] = "General";
ExecuteAFile (DIRECTORY_SEPARATOR + "home" + DIRECTORY_SEPARATOR + "macmanes" + DIRECTORY_SEPARATOR + "hyphy" + DIRECTORY_SEPARATOR + "HYPHY" + DIRECTORY_SEPARATOR + "TemplateBatchFiles" + DIRECTORY_SEPARATOR + "ModelSelectorCodon.bf"); 



I always get this error:
Code:
Log Likelihood = -2999.54815749801;
Shared Parameters:
NSR0=1.05924
G3=0.549061
G2=0.313017
G1=0.446134
C3=0.317419
C2=0.266713
C1=0.22303
A3=0.229822
A2=0.391362
A1=0.273183
AC=0.295982

Tree codon_tree_1=[deleted for the purposes of this post..];
{
{     0.27318347597,    0.391362301614,    0.229822129676}
{    0.162101866427,    0.162331788349,    0.244469117074}
{    0.251938683941,    0.139701221038,    0.288646342406}
{    0.312775973661,    0.306604688999,    0.237062410844}
}

1.13792

Base model has 180 parameters
Log Likelihood = -2999.54815749801;
Shared Parameters:
NSR0=1.05924
G3=0.549061
G2=0.313017
G1=0.446134
C3=0.317419
C2=0.266713
C1=0.22303
A3=0.229822
A2=0.391362
A1=0.273183
AC=0.295982

Tree codon_tree_1=[tree, deleted to this post more readable];
Error:
Invalid Matrix Index [0][0] in a -2 by 2 matrix.
Call stack
1 : sortedScores[0][0]=-crapAIC

Segmentation fault
 



Back to top
 
 
IP Logged
 
Sergei
YaBB Administrator
*****
Offline


Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male
Re: ModelSelectorCodon.bf bug?
Reply #1 - Aug 10th, 2010 at 12:05pm
 
Hi Matt,

This analysis must be run in an MPI environment; sorry for the lack of meaningful error messages.

Sergei
Back to top
 

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
WWW WWW  
IP Logged
 
Matt
YaBB Newbies
*
Offline


Feed your monkey!

Posts: 23
Re: ModelSelectorCodon.bf bug?
Reply #2 - Aug 10th, 2010 at 1:21pm
 
Hi Sergei,

The thing is, with the HYPHYMPI, the program appears to be non-responsive for hours, at which time I grow impatient and kill it.  This is using the .bf that I included in the original post. I'm running GARD and GABranch and a few other things using MPI, so I'm 100% sure that the MPI environment is OK.

The same .bf, using HYPHYMP, proceeds through several steps (again, copied in original post) within minutes, before it dies. 

So, is this a case where the screen output is different between the MPI and MP versions, or is there something else going on? 

If the former, you might consider, for future HYPHY release, adding a status indicator or some sort of output that would allow the used to have confidence that the program is working as intended.. Most of the algorithms already have such an output. 

Thanks!
Back to top
 
 
IP Logged
 
Sergei
YaBB Administrator
*****
Offline


Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male
Re: ModelSelectorCodon.bf bug?
Reply #3 - Aug 10th, 2010 at 1:25pm
 
Hi Matt,

You should get some feedback on the progress pretty quickly (almost immediately for the first few steps), but this IS a slow analysis (a few hours at least). How big is your alignment? I'll also check the settings in your run to confirm that the code isn't simply hanging and waiting for input.

Thanks for your feedback!

Sergei
Back to top
 

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
WWW WWW  
IP Logged
 
Matt
YaBB Newbies
*
Offline


Feed your monkey!

Posts: 23
Re: ModelSelectorCodon.bf bug?
Reply #4 - Aug 10th, 2010 at 1:33pm
 
85 sequences of length 300bp. 

I get no screen output from this analysis, simply a blinking cursor.. In other analyses, when I screw up the batchfile by forgetting to add something or misspelling an input option, HYPHYMPI usually just seg faults..
Back to top
 
 
IP Logged
 
Sergei
YaBB Administrator
*****
Offline


Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male
Re: ModelSelectorCodon.bf bug?
Reply #5 - Aug 10th, 2010 at 1:41pm
 
Hi Matt,

Thanks; this behavior is not normal. I will check if I can reproduce the issue on my installation with the settings that you selected.

Sergei
Back to top
 

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
WWW WWW  
IP Logged
 
Matt
YaBB Newbies
*
Offline


Feed your monkey!

Posts: 23
Re: ModelSelectorCodon.bf bug?
Reply #6 - Aug 29th, 2010 at 5:10am
 
Hi Sergei,

Any update on this issue?

Thanks, Matt
Back to top
 
 
IP Logged
 
Sergei
YaBB Administrator
*****
Offline


Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male
Re: ModelSelectorCodon.bf bug?
Reply #7 - Aug 29th, 2010 at 3:33pm
 
Hi Matt,

Sorry about the delay -- your question slipped through the cracks. I will take a look tomorrow and see what I can find.

Sergei
Back to top
 

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
WWW WWW  
IP Logged
 
Sergei
YaBB Administrator
*****
Offline


Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male
Re: ModelSelectorCodon.bf bug?
Reply #8 - Aug 31st, 2010 at 12:37pm
 
Hi Matt,

Try adding 'inputOptions' as the second argument to ExecuteAFile (as shown in the quoted code below) -- without this option the analysis will not use your presets as input, and running HYPHYMPI in an MPI will probably simply hang waiting for stdin. I tried running your input on a random codon file with ~50 sequences and all went well.

Also, the recommended option for inputOptions ["08"] is "BIC" (as "c-AIC" leads to some overfitting).

Sergei

Matt wrote on Aug 9th, 2010 at 5:43pm:
Using either HYPHYMPI or HYPHYMP....
Code:
inputOptions={};
...

ExecuteAFile (DIRECTORY_SEPARATOR + "home" + DIRECTORY_SEPARATOR + "macmanes" + DIRECTORY_SEPARATOR + "hyphy" + DIRECTORY_SEPARATOR + "HYPHY" + DIRECTORY_SEPARATOR + "TemplateBatchFiles" + DIRECTORY_SEPARATOR + "ModelSelectorCodon.bf",inputOptions); 


[/code]


Back to top
 

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego
WWW WWW  
IP Logged