Check behavior when N_centers changes between continuing runs

[carolyner]

Check re-evaluation of centers, line 1131, when restarting with N_center less than previously specified (last line below):

    # Recalculate centers
    center_rows <- (n_draw + 1):(n_draw + N_centers)
    iter_parm_draws[center_rows, (all_parm_names)] <- as.data.table(center_next)

Observed error in line 113: Supplied 6 items to be assigned to 3 items of column 'ar.mean'.
Occurred when restarting from a run with N_centers=10 with N_centers (accidentally) set to 3.

In general, the actual number of centers can be less than N_centers (e.g., at the beginning when there are fewer than N_centers points). This error may be specific to re-evaluation of centers when continuing with N_centers less than previously specified, or it may occur because N_centers is less than observed at the previous step, which would generally not occur.

Suggested interim fix: do not allow N_center to be less than previously specified.
Longer-term fix, if this is causing an error: we would need some type of special case when continuing from a prior run but reducing the number of centers.

[cmaerzluft]

I have not been able to reproduce this issue. I did add the change that N_center is automatically set to be N_center from the previous run.

[carolyner]

Ideally, users would be able to reduce the number of centers across runs. This would allow an initial exploration with a large number of centers early, shifting to a more focused draws later. Similarly, it may be desirable to increase the number of centers, particularly to take advantage of parallel computing capabilities. This is would be an enhancement.

The main potential for problems occurs because of how centers are passed for recalculation of targets and re-evaluation of centers. Center information is contained in the N_centers rows after n_draws in data tables iter_parm_draws, iter_sim_target, and iter_target_dist and affect the number of rows when these are initialized.

When the number of centers is decreased from N_1 on a previous run to N_2 on a a continuation run, we would only keep the first N_2 centers in these files; these are the centers that produced simulated targets that are nearest to observed.