Floating point exception (core dumped)

[EB Wong]

Hi everyone,

I am running modeltest-ng to get the best-fit model for raxml-ng. Out of the 30 genes in my file, one of them was aborted half way in the run and received error "Floating point exception (core dumped)" from modeltest-ng. 

Here is the command I was using:

modeltest-ng -d aa -i C0256.laln -o C0256.laln.mt -ff

I saw previous discussion suggested to use only single core for the run (I tried -p 1), but the error still persisted in my case.

I tried a few parameter options on the run and it happened to stop when it is supposed to calculate model+I+G4. As according to the manual, modeltest-ng was considering both the proportion of invariant sites (I) and gamma rate categories (G) in searching the best model with optimum rate heterogeneity by default. So I tested a few times to include I and G option separately in the run. Surprisingly, my file was able to run properly.

However, I would prefer to obtain the best-fit model by including the calculation of both I and G for my sequences or at least understand further on what was happening in the run before opting to either I or G (just in case I have to). Floating point exception sounds something like zero division error to me. Does anybody have any idea on the error and what I should do about it?

Thanks!
Bhei

[Diego Darriba]

Hi Bhei,

The bug related to the multi-threaded version was already fixed. I think I would need to take a look at your data. Could you please send me the input files?

Best,
Diego

--
You received this message because you are subscribed to the Google Groups "ModelTest-NG discussion group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to modeltest-ng...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/modeltest-ng/1ee7779c-12f4-40b3-b0d5-146728ddfb44n%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[EB Wong]

Hi Diego,

Thanks for your reply! I have sent the requested input file to your private email.

Also, I reran the program and the error is gone!

However, it seems that there are huge differences between AICc and both AIC/BIC scores in the output (as in Example). The same case happened for a few other sequence files that weren't so on my previous run, particularly on short sequence files. The error "MSA has too few sites compared to the number of taxa and results might not be reliable" was printed in the output files too. Wondering if different measures/restriction have been applied to such short sequence files and the others?

Example:

The output before using updating ModelTest-NG:

Summary:

Partition 1/1:
                         Model         Score        Weight
----------------------------------------------------------
       BIC             LG+I+G4    25938.5567        0.7392
       AIC             LG+I+G4    25077.7081        0.9320
      AICc            LG+I+G4    25077.7081        0.9320

Execution results written to C0093.laln.log

The output using ModelTest-NG with latest update (downloaded using git clone):

Summary:

Partition 1/1:
                         Model         Score        Weight
----------------------------------------------------------
       BIC             LG+I+G4    25938.5831        0.7368
       AIC             LG+I+G4    25077.7345        0.9312
      AICc              LG+G4   173594.9462        1.0000

Execution results written to C0093.laln.log2

Regards,

Bhei

[Diego Darriba]

Hi Bhei,

I observed that the sample size in your data set (number of sites) is very close to the number of parameters (branch lengths + rate heterogeneity parameters). In such a case, the correction term applied in AICc is very large compared to the AIC score. Probably the crash you observed previously was caused by a division by zero (when n - k - 1 = 0, being n the sample size and k the number of parameters). In a recent fix I prevented this error, but AICc still does not perform correctly for such a small sample size [1]. I will add a warning for this situation in the AICc results section.

[1] Susko, E., & Roger, A. J. (2020). On the use of information criteria for model selection in phylogenetics. Molecular biology and evolution, 37(2), 549-562.

To view this discussion on the web visit https://groups.google.com/d/msgid/modeltest-ng/f29fd422-1c51-4640-b70a-840f4a55d958n%40googlegroups.com.

[Muhammad Faizan Latif]

can anyone please send me the zip file of jmodel test i am trying to download it from github.com the install file don't work i will be thank full to you 

[Muhammad Faizan Latif]

 

ml tree k80+i+g is invalid please check the phyml log tab or run phyml alone ? how to solve this issue kindly help me..

You received this message because you are subscribed to a topic in the Google Groups "ModelTest-NG discussion group" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/modeltest-ng/zvh6x7mnuqg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to modeltest-ng...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/modeltest-ng/57961337-01f8-4400-9ffd-a60976c4ae81n%40googlegroups.com.

[Diego DX]

Hi Muhammad,

Please open a new thread whenever your question is not related to a previous one and set a descriptive subject. It will be more helpful for other users. In order to help you, I would need to see the text in the PhyML log tab. However, I suggest you to use ModelTest-NG instead of jModelTest, which does not depend on PhyML anymore.

Best,

Diego.