Cobaya version issues

[nnedelko]

I have successfully (I think) installed Cocoa, but I'm running into errors trying to run some of the example .yaml files with Cobaya.
Several likelihoods produce errors like [input] *ERROR* likelihood 'planck_2018_lowl.TT' does not recognize some options: 'path' (did you mean 'stop_at_error'?). Check the documentation for 'likelihood'., and some likelihoods can't be found, like sn.roman_o and des_y3.combo_3x2pt (I have DESY3 enabled and the projects/des_y3 folder is present).
I think this may be caused by the fact that the version of Cobaya that gets invoked is the one in my .local/lib/python3.10/site-packages/ rather than the one installed in the cocoa env. Is there a way to rectify this without uninstalling Cobaya from my .local version of Python?

[vivianmiranda]

let me look into that.

I think you may be right because we do patch cobaya automatically. Let me install cobaya in my own .local and see if I can tweak start_cocoa.sh to avoid that issue

[nnedelko]

I've uninstalled the .local version of Cobaya for now. This cleared the [input] *ERROR* likelihood 'planck_2018_lowl.TT' does not recognize some options: 'path' error and the sn likelihoods now work, but the DES Y3 likelihoods still cannot be found despite me both including the DES Y3 project in the earlier Cocoa install and having run the DES Y3 compilation script separately just in case: CosmoLike/cocoa_des_y3#8

[nnedelko]

the DES Y3 likelihoods still cannot be found

Ah. There is seemingly a typo in the example .yaml files. They are referring to the likelihood as des_y3.des_3x2pt whereas the actual likelihood name is des_y3.combo_3x2pt. After fixing this typo everything works. Is this just a mistake or is there some sort of alias thing that didn't kick in on my install?

[nnedelko]

Okay, Cocoa seems to work fine now. Again, so as not to spam Issues, I'll ask a question here: what settings should I run CLASS with if I want to use it instead of CAMB in the DES Y3 Cocoa pipeline?

[vivianmiranda]

ops - we made some changes in the python file names and forgot to propagate

[vivianmiranda]

run CLASS with if I want to use it instead of CAMB in the DES Y3 Cocoa pipeline?

good question. I am not an expert in CLASS

[nnedelko]

Ah, I see. Could you maybe point me to someone from the CosmoLike dev team I can contact about the implementation of CLASS in the DES pipeline?

[vivianmiranda]

Running class is a bit tricky in Cobaya.

We do provide a Planck example on examples/

but even that I can only run in LCDM. We need some expert in Class-Cobaya to help us

[nnedelko]

The Planck example is unfortunately not relevant here because LSS data requires a much higher level of precision from CLASS than CMB likelihoods.

We need some expert in Class-Cobaya to help us

From looking into the cosmolike_core repository it appears that the version of CLASS used there is somewhat modified, but it isn't totally clear in what ways and there is no python wrapper supplied with it, which is why I think I need someone specifically from the CosmoLike team who knows about the state of CLASS in the pipeline. Should I open an inquisitory issue at cosmolike_core or is that repository checked too rarely for that to make sense?

[vivianmiranda]

The cosmolike we use on cocoa is cocoa_cosmolike_core repository

[nnedelko]

cocoa_cosmolike_core

If I understand correctly, that version doesn't have an integrated implementation of CLASS, right? I'd like to find out the details of the CLASS set-up from the earlier Cosmolike pipeline used to prepare the DES Y3 3x2pt paper.

[nnedelko]

At the moment when trying to run DES Y3 Cocoa with CLASS I'm running into an occasional error of the type Not possible to extrapolate to k=1.0232929922807536e-05 1/Mpc (minimum k possible is 1.0306882685963715e-05 1/Mpc)., which is I think raised by CosmoLike? If you've at some point encountered this when using CAMB, what precision parameters would need to be increased in CAMB to fix this?

UPD: actually it's a cobaya/theories/cosmo/boltzmannbase error, but still should equally apply to CAMB.