Failure on gtapp

[jballet]

#Describe the bug
We have encountered a very basic bug that happens at setup level when fermipy tries calling gtselect via gtapp.
The error message is
2026-06-05 12:14:45 INFO GTBinnedAnalysis.run_gtapp(): Running gtselect.
Traceback (most recent call last):
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermitools/GtApp.py", line 52, in _test_par_value
self.pars[key]
~~~~~~~~~^^^^^
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermitools/pil.py", line 62, in getitem
return float(value)
^^^^^^^^^^^^
ValueError: could not convert string to float: 'np.float64(100.975)'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/dsm/saplxglast/home/jbb/Flight/14years/test/PbGTASetup/./pycmd.py", line 9, in
status = gta.setup()
^^^^^^^^^^^
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermipy/gtanalysis.py", line 1096, in setup
c.setup(overwrite=overwrite)
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermipy/gtanalysis.py", line 5184, in setup
self._select_data(overwrite=overwrite, **kwargs)
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermipy/gtanalysis.py", line 5265, in _select_data
run_gtapp('gtselect', self.logger, kw_gtselect,
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermipy/gtanalysis.py", line 217, in run_gtapp
gtapp[k] = v
~~~~~^^^
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermitools/GtApp.py", line 44, in setitem
self._test_par_value(key)
File "/dsm/manip/glast/anaconda3/envs/fermi-2.5.1/lib/python3.11/site-packages/fermitools/GtApp.py", line 55, in _test_par_value
raise ValueError("for parameter %s, %s" % (key, ee.args[0]))
ValueError: for parameter ra, could not convert string to float: 'np.float64(100.975)'

This did not occur with fermipy 1.4.0.

To Reproduce

PbGTASetup.zip

Here is a very basic test harness. The ltcube file is fake, but the script fails before it needs it.

Environment (please complete the following information):

• OS: 5.14.0-611.55.1.el9_7.x86_64
• fermipy 1.4.1
• fermitools 2.5.1

[facero]

Hi Jean
I wonder if this is similar to a problem I encountered recently.
In my case it was related to numpy version.
It seems fermipy is not yet compatible with numpy >2.
Can you check wich numpy version you have ?

In my case forcing to an older numpy fixed it.
pip install numpy=1.26.4

But this is a more general issue that should be adressed

[jballet]

Thank you, Fabio.
Indeed I see numpy 2.4.4 whereas at SLAC numpy 1.26.4 is used.

[facero]

For the moment the environment yaml file does not requires numpy < 2.0 but the code fails with any numpy > 2.

the issue here I think is that in numpy 2.x, repr(np.float64(100.975)) returns the string np.float64(100.975) instead of 100.975

But I fear this also requires a fix in fermitools not only fermipy.

Maybe somewhere in the GtApp.py file here :
https://github.com/fermi-lat/sane/blob/719df17835031d6999340c86c278ad18e6fe3d6d/python/GtApp.py#L55

[ndilalla]

Thanks @facero! Indeed fermitools is not yet compatible with numpy >= 2, so we never tested fermipy with recent versions of numpy.

@jballet can you test if downgrading to numpy < 2 solve this issue? If so, I will add this to the requirements

[facero]

I think the changes to fermitools will be pretty minimal to be compatible with numpy 2. In the short term this will be a problem as gammapy just dropped python 3.11 and numpy 1.26 will not be supported in future python version. And gammapy is a dependency of fermipy. I had a test version with a fixed numpy 2 version of feemitools and it's just a couple of lines. Where should I open a PR on the fermitools sane repo ? Le mar. 9 juin 2026, 18:21, Niccolò Di Lalla ***@***.***> a écrit :

…

*ndilalla* left a comment (fermiPy/fermipy#675) <#675 (comment)> Thanks @facero <https://github.com/facero>! Indeed fermitools is not yet compatible with numpy >= 2, so we never tested fermipy with recent versions of numpy. @jballet <https://github.com/jballet> can you test if downgrading to numpy < 2 solve this issue? If so, I will add this to the requirements — Reply to this email directly, view it on GitHub <#675?email_source=notifications&email_token=ACCAQYVEXVWNHMKOIU37SDT47BBRLA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRWGIZDGMJTGQ3KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4662231346>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACCAQYVSOCDH5EU47ZB27PD47BBRLAVCNFSM6AAAAACZ3V3PUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DMNRSGIZTCMZUGY> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/ACCAQYQKN4EBL6WKADCUCV347BBRLA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRWGIZDGMJTGQ3KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJKTGN5XXIZLSL5UW64Y> and Android <https://github.com/notifications/mobile/android/ACCAQYWMCGLC6Q4B5CXAD2T47BBRLA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRWGIZDGMJTGQ3KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI>. Download it today! You are receiving this because you were mentioned.Message ID: ***@***.***>

[jballet]

I am confused by the github organization of the Fermi Tools.
I would not know where to declare that.
Can you share the fix so that we can try it too?

[jballet]

I confirm that downgrading numy to numpy 1 fixes the problem.

[facero]

I describe the fix (just 2 lines) in the fermitools repo issue :
fermi-lat/sane#3

It is not a very active repo and I don't know where I should open a PR. I'm also a bit lost in the github organization repo.
@ndilalla do you know where I can make a PR to fix to np2.
This will be quiet nice to make that upate for the future maintanabilty of fermitools+fermipy.

A patch for fermipy is also needed if you use catalog.py in fermipy which I can open a separated PR .

[ndilalla]

I describe the fix (just 2 lines) in the fermitools repo issue : fermi-lat/sane#3

It is not a very active repo and I don't know where I should open a PR. I'm also a bit lost in the github organization repo. @ndilalla do you know where I can make a PR to fix to np2. This will be quiet nice to make that upate for the future maintanabilty of fermitools+fermipy.

A patch for fermipy is also needed if you use catalog.py in fermipy which I can open a separated PR .

Thanks @facero ! If you can open a PR on fermipy I can review it and merge it. Thanks!