iminuit strategy error when using cabinetry.fit.fit

[TommyDESY]

Hi everyone,

I currently get an iminuit error using versions 0.6.0 of cabinetry and 0.7.6 of pyhf which had never occurred before. Below is the full error thread. model is built with pyhf.pdf.Model, data is a numpy array and minos is a list containing the name of my parameter of interest as a string (the error also occurs when the minos parameter is not set at all). I also tried to set the iminuit strategy manually in cabinetry.fit.fit to ensure that it's an integer but I get the same error. Does anyone know of a fix besides downgrading iminuit to an older version? Thanks for your help.

  File "/nfs/dust/belle2/user/tommy/Vub/inclusive_xulnu/xulnu/pyhf_utils.py", line 890, in extract_signal_strength_scale_factors
    fit_res = cabinetry.fit.fit(model, data, minos=minos)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/nfs/dust/belle2/user/tommy/cabinetry/src/cabinetry/fit/__init__.py", line 367, in fit
    fit_results = _fit_model(
                  ^^^^^^^^^^^
  File "/nfs/dust/belle2/user/tommy/cabinetry/src/cabinetry/fit/__init__.py", line 159, in _fit_model
    result, corr_mat, best_twice_nll, result_obj = pyhf.infer.mle.fit(
                                                   ^^^^^^^^^^^^^^^^^^^
  File "/afs/desy.de/user/t/tommy/.local/lib/python3.11/site-packages/pyhf/infer/mle.py", line 131, in fit
    return opt.minimize(
           ^^^^^^^^^^^^^
  File "/afs/desy.de/user/t/tommy/.local/lib/python3.11/site-packages/pyhf/optimize/mixins.py", line 193, in minimize
    result = self._internal_minimize(
             ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/afs/desy.de/user/t/tommy/.local/lib/python3.11/site-packages/pyhf/optimize/mixins.py", line 52, in _internal_minimize
    result = self._minimize(
             ^^^^^^^^^^^^^^^
  File "/afs/desy.de/user/t/tommy/.local/lib/python3.11/site-packages/pyhf/optimize/opt_minuit.py", line 119, in _minimize
    minimizer.strategy = strategy
    ^^^^^^^^^^^^^^^^^^
  File "/cvmfs/belle.cern.ch/el9/externals/v02-02-01/Linux_x86_64/common/lib/python3.11/site-packages/iminuit/minuit.py", line 228, in strategy
    self._strategy.strategy = value
    ^^^^^^^^^^^^^^^^^^^^^^^
TypeError: (): incompatible function arguments. The following argument types are supported:
    1. (arg0: iminuit._core.MnStrategy, arg1: int) -> None

Invoked with: <iminuit._core.MnStrategy object at 0x7f08daedc830>, {}

[alexander-held]

Hi @TommyDESY, I have not seen this before and I am not aware of any particular issues with the latest iminuit version. Could you please share a way to reproduce this issue for me to have a closer look? Does changing the iminuit version change the behavior and can you pinpoint the version where the behavior changes?

[TommyDESY]

Hi @alexander-held,

here's a minimal example to reproduce the issue:

import pyhf
import cabinetry

spec_test = {
    'channels': [
        {'name': 'signal',
         'samples': [
             {'name': 'Xulnu_in',
              'data': [76, 102],
              'modifiers': [{'name': 'FEI_B0',
                             'type': 'normsys',
                             'data': {'hi': 0.9880450809276028, 'lo': 1.011954919072397}}]}]}]}

model = pyhf.Model(spec_test, poi_name='')
asimov_data = cabinetry.model_utils.asimov_data(model)
cabinetry.fit.fit(model, asimov_data)

I tested this small script with iminuit versions 2.15.1 to 2.26.0 and I always get the exact same error. Again, I use pyhf v.0.7.6 and cabinetry v0.6.0.

[alexander-held]

I cannot reproduce the error with this. I think you probably have a modified local version of cabinetry (picked up from /nfs/dust/belle2/user/tommy/cabinetry) which differs from v0.6.0 and causes the difference. In the traceback

  File "/nfs/dust/belle2/user/tommy/cabinetry/src/cabinetry/fit/__init__.py", line 159, in _fit_model
    result, corr_mat, best_twice_nll, result_obj = pyhf.infer.mle.fit(
                                                   ^^^^^^^^^^^^^^^^^^^

points to a line that is different from where this line is in the released version:

cabinetry/src/cabinetry/fit/__init__.py

Line 94 in 2af877c

result, corr_mat, best_twice_nll, result_obj = pyhf.infer.mle.fit(

Could you please see if any local modification you have done might be responsible for this error?

[TommyDESY]

Yes indeed, it was due to my changes to cabinetry, I should've noticed it. Thanks for the help!