Problem with get_pk_and_k_and_z() in python wrapper

[fmccarthy]

Hi; I am finding that the below code produces the following error in v3.1.2

"classy.pyx in classy.Class.get_pk_and_k_and_z() 
ValueError: negative dimensions are not allowed"

It seems that computing the tCl along with mPk messes something up that is required for get_pk_and_k_and_z()

from classy import Class

params={'N_ncdm': 1,
        'N_ur': 2.0328,
        'non linear': 'hmcode', 
        'output': 'mPk', 
        'z_pk':'3',
        'P_k_max_1/Mpc': 15,
        '100*theta_s': 1.0417368567740997, 
        'A_s': 2.1102839631833067e-09, 
        'm_ncdm': 0.06, 
        'n_s': 0.9671828913167801, 
        'omega_b': 0.02255434170350831,
        'omega_cdm': 0.11817158102801589,
        'tau_reio': 0.05515052267600599}

c = Class()

c.set(params)

c.compute()

print("getting pk")

c.get_pk_and_k_and_z()   # this works fine

print("got pk")

c.set({'output':'mPk tCl'})

c.compute()
print("getting pk")

c.get_pk_and_k_and_z()    # this line produces an error

print("got pk")

[jcolinhill]

Echoing Fiona's comment -- this error breaks (e.g.) the interface between CLASS and any likelihood in Cobaya that requires the matter power spectrum, such as DES. I briefly tried to find a fix, but with no immediate success -- it appears to be indexing-related somehow.

[jcolinhill]

(note that this must be due to a change from CLASS v2 to v3, as this error is not thrown when running P(k)-based likelihoods with CLASSv2 in Cobaya)

[alexander-reeves]

Hi both, I am also encountering the same issue- has there been any update or workaround of this? I have tried installing previous versions of Class but with no success.

Many thanks,

Alex

[svenguenther]

Hey all, I could reproduce that bug. For me it works that I downgrade this function to the version 3.1.1. It is communicated to the maintainer of CLASS and will be fixed in one of the next releases. Thank you for making us aware of that problem. Best, Sven

[schoeneberg]

Dear all, this has been finally propagated to me, and the solution is now in the newest development branch and will be released as a minor release shortly!

[jcolinhill]

Thanks @schoeneberg !

[fmccarthy]

Thanks for sorting that out @schoeneberg !

[knaidoo29]

Hi, I've come across the same error. Is the update coming anytime soon? What can I do to get this working in the mean time? I've added a model to class so can't move to an older version.

[schoeneberg]

Dear @knaidoo29 , the PR on this issue is already underway on the private devel branch, and should be released with the next minor release. When that will be, I cannot tell you. You can write an email to @lesgourg and ask him about it :smile:

I think there's no easy "working-in-the-meantime" fix, at least not one I could describe in a comment. It might help to include tCl/pCl in your output if you haven't already, but beyond that, I do agree that the PR is urgent. Feel free to contact Prof. Julien Lesgourgues about this issue if it is blocking your work.

[kcroker]

@schoeneberg is this fixed in v3.2.1? If so, could we close this issue?

[schoeneberg]

Dear all, I confirm that the bug is fixed and the fix is propagated to v3.2.1. ( @kcroker , @knaidoo29 , @fmccarthy , @jcolinhill , @alexreevesy ) Sadly I cannot currently close any issues, so this will have to be done from @lesgourg !

[kcroker]

Dear all, I confirm that the bug is fixed and the fix is propagated to v3.2.1. ( @kcroker , @knaidoo29 , @fmccarthy , @jcolinhill , @alexreevesy ) Sadly I cannot currently close any issues, so this will have to be done from @lesgourg !

@schoeneberg thank you for following up quickly. @fmccarthy as you opened this issue, I think you also have the ability to close it? If so, could you please close it?

[fmccarthy]

thank you for fixing this!