ValueError: math domain error

[matthewglasenapp]

Hello,

When running QuIBL, I encounter a "ValueError: math domain error" after the program runs for quite some time. I've tried two different data set sizes (16801 trees, 7645 trees). I've also varied the numsteps parameter from 50 - 200. The gene trees were reconstructed with iqtree using the following command: iqtree2 -S <directory_of_fasta_alignments>

I made sure that I did not include trees with 0-length branches. I'm running the program on an Intel 24 core node with 128 GB memory capacity. Any tips would be greatly appreciated.

Best, Matt

error_log.txt trees2.txt InputFile.txt

[stubbsrl]

Hello,

Has there been on progress on this? I looked through the closed and open issues and the "math domain error" is quite common and problematic. I split my 3000+ trees with 19 tips into two groups and I was able to run them successfully. So, I am sure it isn't an issue with a single tree. Also, I have also had the error for running out of memory, so I know I have the resources properly allocated (and I ran quibl successfully with 5000+ trees on these same resources at one point).

@matthewglasenapp can I ask why you varied the numstep parameter and what if any affect that had? Did you try splitting your trees into two groups?

Any help is appreciated!

Thanks, Rebecca

[laufran]

I just wanted to bump this error as I've been having it as well. I'm running QuIBL on 1000 gene trees with 10 tips generated from Seq-Gen. I've confirmed that there are no zero-length edges on all the gene trees. I've attached my gene trees file, the QuIBL command file, and the exact error message I get after a variable amounts of triplets have already been successfully analyzed. I've tried this for a variety of independent gene tree files and this happens for nearly all of them.

gts.txt quibl.txt error.txt

[jihoonjo]

Hello, I also met the math domain error, It seems an error from during the log transformation of the '0' value. So I manually fixed it with a minor correction of the 'Quibl.py' python script. And, it worked well.

The fixed line is line 139 in 'QuIBL.py', like below.

before: L+=log(temp)

after (you have to appropriately align the tab or space): if temp != 0: L+=log(temp)

I hope it will be helpful to your research.

I attached the fixed code. QuIBL.py.txt

[cecileane]

but log(0)= - infinity so is it appropriate to pretend as if log(temp) = 0 when temp is 0? I understand this fix avoids the technical error, which is great. I wonder if it introduces a mathematical error in the calculation of the criterion.

[Dived-Jin]

Hello, I also met the same problem when I randomly selected 500 1-Kb trees constructed by iqtree2. I also confirmed that don't have a branch length with zero in my data sets. But most replicates end with the same error at QuIBL.py line 139 (log(temp)). So has anyone solved this problem?

[juanenciso14]

If you guys still have this problem, maybe you could try what I did. After facing this problem with the original script that is in the main folder, I managed to consistently run batches of 500 trees with 43 tips each, using the Cython version and maxcores:2. I'm not sure which of the two did the trick but this way I managed to overcome this error.

[xiaolongtu]

hello, i also meet this, have anyone solve this probelm ?

[juanenciso14]

hello, i also meet this, have anyone solve this probelm ?

Hey @xiaolongtu! Check my comment above, it may work for you as well.

[xiaolongtu]

@juanenciso14 , thank you , I hvae try it but still has wrong