high-dimension models

[LydiaMak]

Hi,

Does anyone have tips to help when running nested sampling for higher dimension models e.g. alpha (most preferred in a quantitative way).

Also, what time frame should I expect for a personal MacBook?

I tried the nlive_init/batch increased to 1000 but didn't seem to help that much.

Thank you,

Lydia

[jrleja]

Hi,

I have used the following settings for recent Ndim ~ 15 fits:

run_params['dynesty'] = True
run_params['nested_weight_kwargs'] = {'pfrac': 1.0}
run_params['nested_nlive_batch'] = 300
run_params['nested_nlive_init'] = 350 
run_params['nested_dlogz_init'] = 0.02
run_params['nested_maxcall'] = 7500000
run_params['nested_maxcall_init'] = 7500000
run_params['nested_sample'] = 'rwalk'
run_params['nested_maxbatch'] = None
run_params['nested_target_neff'] = 20000
run_params['nested_first_update'] = {'min_ncall': 20000, 'min_eff': 7.5}

Fits typically take ~10-15 hours per object for photometry, with the runtime being dominated by model generation.

If you're looking to get quicker fits, I'd suggest looking into the recently-released neural net emulator for Prospector-alpha: https://ui.adsabs.harvard.edu/abs/2023arXiv230616442M/abstract. This cuts the runtime down to ~10 minutes / object.

Best,
Joel

[LydiaMak]

Hi,

This is great! Thank you a lot! I will try that and see how it's going! Thank you again for the quick reply!

Best,
Lydia