Continue and parallel computation about sisso HOT 3 CLOSED

1. Yes.

2. I guess that you have small jobs so that 2 nodes take same time with that using 1 node, right? Nevertheless the code indeed need to be optimized by a computer expert, and this is in progress.

from sisso.

Thanks for your reply.

1. However I always encounter a mpirun error when I try to restart some runs. And I've checked it is not the memory problem.
2. The job took 4 hours (I only tested using rung = 2), it took exactly the same time using 2nodes vs 1node.
3. Have you tested how much a difference rung=2 vs rung=3 would give to the result? And also the subs_sis? Like does it have to at least above some value then it would give any reliable results. In a script you send me I see you used 10k, whereas for my problem, rung=3 and subs_sis=10k take too long. Based on some of my tests, subs_sis=1k, 5k, 10k can generate the same 1D descriptor.

Thanks!

from sisso.

I need to see your input setting before I can answer your questions. So please send me an email if that is OK. rung=2 and rung=3 should give different results since rung=3 corresponds to bigger feature spaces with more complex features. In principle, the larger the subs_sis the higher the probability to identify the best model. So I would use as large subspace as possible unless e.g. I see a convergence. The 1D is the same no matter what subs_sis was set since the 1D is just the first feature selected by SIS.

from sisso.