Resuming from Last Run

mica2 supports resuming from last run. The adopted DNest sampling code CDNest (see Diffusive Nested Sampling) outputs a number of sampling snapshots with a name as restart_dnest.txt_xxx for continuum reconstruction and restart_dnest1d.txt_yy_xxx for time delay analysis (mica2 first reconstructs the continuum light curve to determine the best values of parameters for the damped random walk model and uses these values as priors for subsequent time delay analysis.). Here, xxx mean the steps when the snapshot is saved and yy mean the number of Gaussians.

The default steps for continuum reconstruction are 2000, so the corresponding snapshots name is restart_dnest.txt_2000. The default steps for time delay analysis are 2000 too, so the corresponding snapshots name is restart_dnest1d.txt_yy_2000. If one sets different steps, the snapshot name is changed accordingly.

Rename the above files as

cp restart_dnest1d.txt_yy_xxx restart_dnest1d.txt_yy     # namely, remove "_xxx"

Then change the step numbers in the param file and resume from last run as

mpiexec -n np ./mica2 para/param -r

where “np” is the number of cores. It must be the same as the number of cores used in last run.

In the Python verson pymica, the function prepare() copies the snapshots and the function restart() resumes from last run. See the Python scripts the subfolder tests/restart in the source package.