Closed esadri21 closed 5 years ago
There's not really enough information here for me to see what's going on, but given the setup my first question is that you have ndim=2
here but most of your functions seem to be just functions of 1 parameter (H0). Is that intended? Right now you're sampling 2 numbers, transforming them from [0, 1]
-> [50, 80]
, feeding both of these into loglike
-> SN
-> dL
, computing 2 values of h
, and likewise returning 2 values for fn
.
found the answer. Oh god. It was about log10
and np.log10
Could you tell me why is very faster than
sampler = NestedSampler(loglike, prior_transform, 3)
this one?
dsampler = dynesty.DynamicNestedSampler(loglike, prior_transform, ndim=3)
Which one is more accurate? Of course Ii think I should use the multiprocessing, because for 5 dimensional models it lats for a loooooong time. I appreciate your attention
The normal nested sampler in general is faster. You’re more than welcome to try to use multiprocessing; dynesty supports such efforts through a user-provided pool.
Dear J. Speagle. For testing the Dynesty, I just copied and pasted my Lambda CDM code form EMCEE into Dynesty. I used the
Loglik
as I used in EMCEEI copied and pasted the relevant part here
the total loglike for different data will be like
SN() + BAO() + ...
But here we just have SN.Now, This code is completely correct in EMCEE and Multinest but when I use it in Dynesty, I get many errors.
sampler = NestedSampler(loglike, prior_transform, 2)
I vectorized
SN()
function and inloglike
I used().any()
. it then reached the answer but wrong values. I am confused what should I do.