'std::length_error' bug? about dsuite HOT 6 CLOSED

Hello David

Thank you for this report. The error occurred because the default jackknife block size for calculating the p-values was 20,000 SNPs and your dataset contained only ~3,000 sites. Therefore the p-values could not be calculated and the program terminated with an error. The solution would be to reduce the block size to something like 100 SNPs or even 50 SNPs (using the -j option).

In response to your report, I implemented exception handling for this, so that, instead of crashing, the program informs the user what the issue is, like below:

WARNING: Not enough blocks to calculate jackknife!!
Could not calculate p-values for the trio: A_calliptera Alticorpus_geoffreyi Copadichromis_cf_trewavasae
You should probably decrease the the jackknife block size (-j option)

...
p-value could not be claculated for 62196 trios
You should definitely decrease the the jackknife block size!!!

from dsuite.

Dear Milan,

Thank you very much for your super fast response and for implementing the warning on this issue! Very much appreciated!

Just as additional information for other users: I think you can also run into this issue even though you have a lot of SNPs in your dataset, in case that you have one individual (or one group with individuals) containing a lot of missing data, so that in the end for the jackknife procedure, less than -j sites remain for the particular comparison(s) involving that individual.

Best wishes,
David

from dsuite.

Yes David, that's correct, the number of sites available for calculating D is different for each trio, due to different amounts of missing data and also other factors - e.g. sites where all individuals in the trio have the same allele are not used. The program will write a warning about any trios for which the p-value can't be calculated due to insufficient number of useable SNPs. The p-value in the output files then appears as 'nan'.

from dsuite.

As said, the solution is to use a lower value for the -j option...

from dsuite.

Yes indeed, thank you for confirming that. I am looking forward to use your tool for many future datasets! Thanks again.

from dsuite.

You're welcome. I will be happy to hear any future feedback, feature requests, etc.

Milan

from dsuite.