search

ACTCollaboration / pyactlike

ACTPol CMB power spectrum likelihood in Python
MIT License
6 stars 6 forks source link

Inf value for Loglkl #17

Open ash2223 opened 2 years ago

ash2223 commented 2 years ago

Hello! I'm trying to run pyactlike with MontePython on my cluster, but i'm getting -LogLkl = inf in my chains:

Screen Shot 2022-08-25 at 6 00 01 PM

I went into the init.py file and found that the analysis never makes it past the line cl = self.get_cl(cosmo, 6000) before it throws an error and sets Loglkl to inf. I'm wondering if I could get some help on bypassing this issue? Thank you!

-Adam

ash2223 commented 2 years ago

Ah, I figured out what the issue was! Turns out "data.cosmo_arguments['l_max_scalars'] = 6000" must be added to the .param file, otherwise the default value that CLASS uses for l_max_scalars is zero, and the get_cl function won't work. Perhaps someone could add this line to the .param file so that this issue will be resolved for others who run into it. Thanks all!

-Adam

shan1525 commented 2 years ago

Dear @ash2223 and @xzackli ,

I am running the MontePython interface for ACT pol_lite likelihood. The param file that is attached with the likelihood gives inf -Loglkl. So as suggested in this thread I used "data.cosmo_arguments['l_max_scalars'] = 6000". The error was resolved but the -loglkl is continiously increasing and the chains are not converging. So as raised in #12, is there any starting covmat and bestfit files. Or is their anything more to get a convergence? Can anybody suggest any solution to this problem. I am using only the act_likelihood with the tau_prior likelihood as given in param file.

with the best regards, shan.

shan1525 commented 2 years ago

Dear developers,

Any update on this. Please, respond whenever you are free.

Thanking you, with the best regards, shan.

xzackli commented 2 years ago

I'm sorry, distracted with some other stuff. Let me get back to you in a few days!

shan1525 commented 2 years ago

I am extremely sorry to disturb you at your busy schedule. Please, take your time. I will wait, whenever you are free, please look into it.

With the best regards, Shan.

On Thu, 27 Oct 2022 at 7:45 PM, Zack Li @.***> wrote:

I'm sorry, distracted with some other stuff. Let me get back to you in a few days!

— Reply to this email directly, view it on GitHub https://github.com/ACTCollaboration/pyactlike/issues/17#issuecomment-1293591918, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQWQH5GJ6CP7DHETMFD7FOTWFKFBDANCNFSM57VACHIA . You are receiving this because you commented.Message ID: @.***>