thomson_fitting error with lmfit==1.3.0

[pheuer]

Bug description

I'm looking into this: I can reproduce this error locally by upgrading my lmfit version from 1.2.0 to 1.3.0 (the latter is what PlasmaPy is now running on, at least in CI). The error occurs in thomson_fitting.ipynb:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[9], line 2
      1 fit_kws = {}
----> 2 epw_result = epw_model.fit(
      3     epw_skw,
      4     params=params,
      5     wavelengths=epw_wavelengths.to(u.m).value,
      6     method="differential_evolution",
      7     fit_kws=fit_kws,
      8 )

File ~\anaconda3\envs\dev311\Lib\site-packages\lmfit\model.py:1174, in Model.fit(self, data, params, weights, method, iter_cb, scale_covar, verbose, fit_kws, nan_policy, calc_covar, max_nfev, coerce_farray, **kwargs)
   1168     fit_kws = {}
   1170 output = ModelResult(self, params, method=method, iter_cb=iter_cb,
   1171                      scale_covar=scale_covar, fcn_kws=kwargs,
   1172                      nan_policy=self.nan_policy, calc_covar=calc_covar,
   1173                      max_nfev=max_nfev, **fit_kws)
-> 1174 output.fit(data=data, weights=weights)
   1175 output.components = self.components
   1176 return output

File ~\anaconda3\envs\dev311\Lib\site-packages\lmfit\model.py:1559, in ModelResult.fit(self, data, params, weights, method, nan_policy, **kwargs)
   1557 self.userargs = (self.data, self.weights)
   1558 self.userkws.update(kwargs)
-> 1559 self.init_fit = self.model.eval(params=self.params, **self.userkws)
   1560 _ret = self.minimize(method=self.method)
   1561 self.model.post_fit(_ret)

File ~\anaconda3\envs\dev311\Lib\site-packages\lmfit\model.py:997, in Model.eval(self, params, **kwargs)
    964 def eval(self, params=None, **kwargs):
    965     """Evaluate the model with supplied parameters and keyword arguments.
    966 
    967     Parameters
   (...)
    995 
    996     """
--> 997     return coerce_arraylike(self.func(**self.make_funcargs(params, kwargs)))

File ~\GitHub\PlasmaPy\plasmapy\diagnostics\thomson.py:685, in _spectral_density_model(wavelengths, settings, **params)
    678 """
    679 lmfit Model function for fitting Thomson spectra.
    680 
    681 For descriptions of arguments, see the `thomson_model` function.
    682 """
    684 # LOAD FROM SETTINGS
--> 685 ion_z = settings["ion_z"]
    686 ion_mass = settings["ion_mass"]
    687 probe_vec = settings["probe_vec"]

TypeError: 'NoneType' object is not subscriptable

Raised in #2618

No helpful details in the lmfit release log, although 1.3.0 released yesterday, so that tracks with the sudden change
https://lmfit.github.io/lmfit-py/whatsnew.html#version-1-3-0-release-notes-april-4-2024

[pheuer]

I think this is a bug in lmfit 1.3.0. The documentation states that additional keywords to Model are passed to the model function, but this clearly isn't happening based on the following minimal non-working example. I will open an issue on the lmfit repository

import numpy as np
from lmfit import Model, Parameters

# Define some data
x = np.linspace(0,1,num=10)
data = 10*x+2



# Parameters list with one parameter
params = Parameters()
params.add("b", value=5, vary=True, min=0.5, max=25)


# Dictionary of constants to be passed to model function through **kws
# of `Model`
settings = {'a':2}


# Define a model function that takes the parameters and the settings keyword
def _model(x, settings=None, **params):
    
    # TypeError is raised here because `settings` keyword is not being 
    # passed through to `_model`, so `settings=None`
    a = settings['a']
    b = params['b']
    
    return x*b + a


model = Model(
     _model,
     independent_vars=["x"],
     settings=settings,
 )


# Try to fit
# This raises an exception
result = model.fit(
    data,
    params=params,
    x=x,
    method="leastsq",
)

print(result.best_values)

[pheuer]

If lmfit doesn't want to restore this functionality (or it takes a while to get that in their release version), the following fix would work (around line 1007 of thomson.py). The lambda function provides a work around way of passing settings into _spectral_density_model

_spectral_density_model_lambda = lambda wavelengths, **params: _model(wavelengths, settings=settings, **params)

 return Model(
        _spectral_density_model_lambda,
        independent_vars=["wavelengths"],
        nan_policy="omit",
        )

[pheuer]

Closed by #2623