Hello,
I'm using CLM-FATES at a single site for a multi-instance (ensemble) simulation. I'm curious if its possible to restart a run (CONTINUE_RUN=TRUE) with a subset of specific ensemble members. The goal is to filter ensemble members at an early stopping point to cull members that clearly won't perform well in the longer term and then continue the run with fewer ensemble members to cut down on computational cost. Its seems to work to change ninst to a lower number and then continue the run as long as the ensemble members you want to continue with are sequential starting at 1 (e.g. 0001 to 0010), but what if you just want to continue with say 0013 and 0026 and cull the rest? Does anyone have experience doing something like this? Thanks in advance for any advice!
The workaround that people have used in the past is to just start another ensemble from scratch using only the param sets that passed the initial test and then run them longer in the second case, but this seems less efficient that continuing the run (if possible).
Adam
PS I'm running on Cheyenne
I'm using CLM-FATES at a single site for a multi-instance (ensemble) simulation. I'm curious if its possible to restart a run (CONTINUE_RUN=TRUE) with a subset of specific ensemble members. The goal is to filter ensemble members at an early stopping point to cull members that clearly won't perform well in the longer term and then continue the run with fewer ensemble members to cut down on computational cost. Its seems to work to change ninst to a lower number and then continue the run as long as the ensemble members you want to continue with are sequential starting at 1 (e.g. 0001 to 0010), but what if you just want to continue with say 0013 and 0026 and cull the rest? Does anyone have experience doing something like this? Thanks in advance for any advice!
The workaround that people have used in the past is to just start another ensemble from scratch using only the param sets that passed the initial test and then run them longer in the second case, but this seems less efficient that continuing the run (if possible).
Adam
PS I'm running on Cheyenne