I ran the diablo a couple of months ago and ncomp = 5 and now when I am running diablo I am getting ncomp = 1. I would like to know if ncomp can change here ncomp <- perf.diablo.MB$choice.ncomp$WeightedVote["Overall.BER", "mahalanobis.dist"] without any changes in parameters, datasets and seed value.
The package is updated on bioconductor every 6 months, but I doubt this would have affected the results. We are using repeated cross-validation. It is possible that even when setting the seed, (and also if you choose parallelisation or HPC) that the sampling may still differ. Have you tried rerunning a few times to see if the results change again?
It could also be that your data are ‘unstable’, i.e several solutions (or optimal) exist in your case.
Kim-Anh
Hi Kim-Anh,
Thanks for your reply. I agree that either data may be unstable and also, I used very few nrepeat when training the model first time. I am trying again to develop a robust model.
Varun