LMFit powell/cobyla crash with ftol when a tolerance is set

[rozyczko]

Setting Fitter.tolerance and fitting with any gradient-free LMFit method (LMFit_powell, LMFit_cobyla) makes the fit raise immediately:

easyscience.fitting.minimizers.utils.FitError: minimize() got an unexpected keyword argument 'ftol'

Without a tolerance, the same fit completes, which makes the bug easy to miss. LMFit_differential_evolution is affected by the same keyword-selection logic.

Reproducible with

import numpy as np
from easyscience import AvailableMinimizers, Fitter, ObjBase, Parameter

class AbsSin(ObjBase):
    def __init__(self, offset_val, phase_val):
        super().__init__('sin',
                         offset=Parameter('offset', offset_val),
                         phase=Parameter('phase', phase_val))
    def __call__(self, x):
        return np.abs(np.sin(self.phase.value * x + self.offset.value))

ref_sin = AbsSin(0.2, np.pi)
sp_sin = AbsSin(0.354, 3.05)
sp_sin.offset.fixed = False
sp_sin.phase.fixed = False

x = np.linspace(0, 5, 200)
f = Fitter(sp_sin, sp_sin)
f.switch_minimizer(AvailableMinimizers.LMFit_powell)   # or LMFit_cobyla
f.tolerance = 1e-4                                     # <- the trigger
f.fit(x=x, y=ref_sin(x), weights=np.ones_like(x))
# FitError: minimize() got an unexpected keyword argument 'ftol'