NaN errors for the Bayesian evidence with dynesty

[jordibp99]

Hello.

When trying to find the best-fitting model for radial velocity data I have issues with the error of the Bayesian evidence for some models in wide volume priors. If I fit a linear trend or a linear trend + 1 planet to the data, everything works correctly. But, when I add an additional planet the Bayesian evidence is calculated but the error is NaN. If I considerably shrink the volume prior, the problem dissapears.

I do not know if anyone has found the same problem and has any idea on the solution.

[nespinoza]

Hi @jordibp99,

Thanks for bringing this up! Could you share a minimum working example for this problem so I can take a look at it? Doesn't have to be your data, it can be a mock dataset.

Thanks in advance!

Néstor

[jordibp99]

My code is almost identical to the juliet tutorials: `directory="/home/jordi/Documents/github/" priors_linear_2p = {}

Name of the parameters to be fit:

params_linear_2p = ['P_p1','t0_p1','P_p2','t0_p2','mu_HARPS','mu_HPF',\ 'K_p1', 'ecc_p1', 'omega_p1','K_p2', 'ecc_p2', 'omega_p2','sigma_w_HARPS',\ 'sigma_w_HPF','rv_intercept', 'rv_slope']

Distributions:

dists_linear_2p = ['uniform','uniform','uniform', \ 'uniform','uniform','uniform','uniform','uniform','uniform','uniform','uniform',\ 'uniform', 'uniform','uniform','fixed','uniform']

Hyperparameters

hyperps_linear_2p = [[0,1000],[2458473.75213848,2459025.63776411],[0,1000],[2458473.75213848,2459025.63776411],\ [-10,10], [-10,10],[0,10], [0,1], [0,360], \ [0,10], [0,1], [0,360], [0, 10], [0, 10],0,[0, 0.01]]

Populate the priors dictionary:

for param, dist, hyperp in zip(params_linear_2p, dists_linear_2p, hyperps_linear_2p): priors_linear_2p[param] = {} priors_linear_2p[param]['distribution'], priors_linear_2p[param]['hyperparameters'] = dist, hyperp

dataset_linear_2p = juliet.load(priors = priors_linear_2p, rvfilename=directory+'data.txt', out_folder = directory+'rvs_lineartrend_2planet') results_linear_2p = dataset_linear_2p.fit(n_live_points = 300)` data.txt

Thanks for the quick answer!

[nespinoza]

I checked, and it seems this is a nested sampling issue, not a juliet one. Closing.