|
Post by bettina on Jul 25, 2014 0:40:43 GMT
Hello,
I would like estimate the variation explained by SNPs on each chromosome. Since my data stem from a highly structured population, I have first estimated the first 10 pc on all autosomes and then tried to run reml mrgm.
My command line looks like this: ./gcta64 --reml --mgrm files --pheno phen --qcovar autos.realatedness.matrix.eigenvec --out autos_simultanous --thread-num 10 --reml-maxit 1000;
I obtain the following error:
Error: Log-likelihood not converged (stop after 1000 iteractions). You can specify the option --reml-maxit to allow for more iterations.
I have increase the maxit parameter to 10000 but still get the same result.
Do you have any explanation for this? Is there any recommendation how many iterations I need to do so that the log likelihood will converge?
Thank you very much in advance?
|
|
|
Post by Jian Yang on Jul 31, 2014 12:06:04 GMT
|
|
|
Post by ninjakate on Jul 22, 2017 12:20:42 GMT
Hi - I have the same error result (for a small sample size). Is there a reasonable number of max iterations to increase to without invalidating the model? Many thanks Kate
|
|
|
Post by ninjakate on Jul 22, 2017 12:33:05 GMT
Addendum - it's also notable that when I do exactly the same analysis on 1M variants (chip data) versus 15M variants (imputed data) the model works, with <100 iterations.
|
|