AttributeError: 'GoshGOS' object has no attribute 'energy_shift'

[liuchzzyy]

Hi, Community
I am following the EELS fitting code of @thomasaarholt for Mn_L edge fine structure:
https://gist.github.com/thomasaarholt/9a123a23bb1ec96f9e42ad2e41483100

However it shows: 'AttributeError: 'GoshGOS' object has no attribute 'energy_shift', which I have no idea about it after I checked that part in github. Then I tried to modifying the code while looking the discussions, but it seems I could still not skip the Error.

I uploaded my jupyter file and relative data in the ZIP below. EELS.zip
it is appreciated for any suggestions!

relevant code as below :

%matplotlib ipympl
s_Mn_L = cluster_high_loss.isig[610.0:700.0].deepcopy()
s_Mn_L.add_elements(('Mn','O'))
model_Mn = s_Mn_L.create_model(low_loss=cluster_low_loss, auto_add_edges=False, auto_background=True,)
model_Mn.components

# hs.set_log_level('ERROR')
model_Mn.fit_component(model_Mn.components.PowerLaw, signal_range=(610.0, 630.0), fit_independent=True, only_current=True)
model_Mn.assign_current_values_to_all(components_list = [model_Mn.components.PowerLaw])
# model_Mn.fit_component(model_Mn.components.PowerLaw, signal_range=(610.0, 630.0), fit_independent=True, only_current=False)


# Create and set the 4 components
import exspy as epy
L3 = epy.components.EELSCLEdge("Mn_L3")
L2 = epy.components.EELSCLEdge("Mn_L2")

L3_white = hs.model.components1D.Gaussian()
L3_white.name = "Mn_L3 line"
L3_white.centre.bmin = 640.0
L3_white.centre.bmax = 645.0
L3_white.centre.value = (L3_white.centre.bmax + L3_white.centre.bmin) / 2 # Initial value of the centre. Must be set to initialise the twinning below.
L3_white.A.bmin = 0
L3_white.sigma.bmax=3

L2_white = hs.model.components1D.Gaussian()
L2_white.name = "Mn_L2 line"
L2_white.centre.bmin = 650.0
L2_white.centre.bmax = 655.0
L2_white.centre.value = (L2_white.centre.bmax + L2_white.centre.bmin) / 2
L2_white.A.bmin = 0
L2_white.sigma.bmax=3

L3.onset_energy.twin = L3_white.centre
L2.onset_energy.twin = L2_white.centre

# Add all four to the model
model_Mn.extend([L3_white, L3, L2_white, L2])
model_Mn.components


# The following sort of emulates smart_fit, but it is more appropriate than smart_fit, 
# because smart_fit doesn't know that it should fit the gaussians between L3 and L2

# We fit "from the left", starting with the lowest energy
model_Mn.reset_signal_range()
model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True,) # get good values
model_Mn.assign_current_values_to_all(components_list = [L3_white]) # apply those good values to all pixels
model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True, only_current=False)

# model_Mn.fit_component(L3, bounded=True, signal_range=[636.,646.])
# mmodel_Mnassign_current_values_to_all(components_list = [L3])
# model_Mn.fit_component(L3, bounded=True, signal_range=[636.,646.], only_current=False)

# model_Mn.fit_component(L2_white, signal_range=[L2_white.centre.bmin-3.0, L2_white.centre.bmax+3.0], bounded=True,)
# model_Mn.assign_current_values_to_all(components_list = [L2_white])
# model_Mn.fit_component(L2_white, signal_range=[L2_white.centre.bmin-3.0, L2_white.centre.bmax+3.0], bounded=True, only_current=False)

# model_Mn.fit_component(L2, bounded=True, signal_range=[649., 659.])
# model_Mn.assign_current_values_to_all(components_list = [L2])
# model_Mn.fit_component(L2, bounded=True, signal_range=[649., 659.], only_current=False)

# # Repeat this for each component
# model_Mn.fit_component(L3, bounded=True, signal_range=[636.,646.], only_current=False)
# model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True, only_current=False)
# model_Mn.fit_component(L2, bounded=True, signal_range=[649., 659.], only_current=False)
# model_Mn.fit_component(L2_white, signal_range=[L2_white.centre.bmin-3.0, L2_white.centre.bmax+3.0], bounded=True, only_current=False)

# model_Mn.multifit(bounded=True) # optional final pass may improve fit by letting all components adjust to each other. 

>   C:\Users\xxx\AppData\Local\miniconda3\envs\txm\lib\site-packages\exspy\components\eels_cl_edge.py:432: RuntimeWarning: invalid value encountered in scalar divide
>   self._power_law_r = math.log(y2 / y1) / math.log(E1 / E2)
> AttributeError                            Traceback (most recent call last)
> Cell In[15], line 6
>       1 # The following sort of emulates smart_fit, but it is more appropriate than smart_fit, 
>       2 # because smart_fit doesn't know that it should fit the gaussians between L3 and L2
>       3 
>       4 # We fit "from the left", starting with the lowest energy
>       5 model_Mn.reset_signal_range()
> -**---> 6 model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True,) # get good values**
>       7 model_Mn.assign_current_values_to_all(components_list = [L3_white]) # apply those good values to all pixels
>       8 model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True, only_current=False)
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\models\model1d.py:890, in Model1D.fit_component(self, component, signal_range, estimate_parameters, fit_independent, only_current, display, toolkit, **kwargs)
>     888     return cf.gui(display=display, toolkit=toolkit)
>     889 else:
> **--> 890     cf.apply()**
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\models\model1d.py:145, in ComponentFit.apply(self)
>     144 def apply(self):
> **--> 145     self._fit_fired()**
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\models\model1d.py:125, in ComponentFit._fit_fired(self)
>     117             self.component.estimate_parameters(
>     118                 self.signal,
>     119                 self.signal_range[0],
>     120                 self.signal_range[1],
>     121                 only_current=only_current,
>     122             )
>     124 if only_current:
> **--> 125     self.model.fit(**self.fit_kwargs)**
>     126 else:
>     127     self.model.multifit(iterpath=self.iterpath, **self.fit_kwargs)
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\exspy\models\eelsmodel.py:553, in EELSModel.fit(self, kind, **kwargs)
>     551     return self.smart_fit(**kwargs)
>     552 elif kind == "std":
> **--> 553     return Model1D.fit(self, **kwargs)**
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\model.py:1649, in BaseModel.fit(self, optimizer, loss_function, grad, bounded, update_plot, print_info, return_info, fd_scheme, **kwargs)
>    1644 # We enforce estimation of the Jacobian if no
>    1645 # analytical gradients available for consistency
>    1646 # with `scipy.optimize.leastsq`
>    1647 auto_deriv = 0 if grad == "analytical" else 1
> **-> 1649 res = mpfit(
>    1650     self._errfunc4mpfit,
>    1651     self.p0[:],
>    1652     parinfo=self.mpfit_parinfo,
>    1653     functkw={
>    1654         "y": self.signal._get_current_data()[
>    1655             self._channel_switches
>    1656         ],
>    1657         "weights": weights,
>    1658     },
>    1659     autoderivative=auto_deriv,
>    1660     quiet=1,
>    1661     **kwargs,
>    1662 )**
>    1664 # Return as an OptimizeResult object
>    1665 self.fit_output = res.optimize_result
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\external\mpfit\mpfit.py:824, in mpfit.__init__(self, fcn, xall, functkw, parinfo, ftol, xtol, gtol, damp, maxiter, factor, nprint, iterfunct, iterkw, nocovar, rescale, autoderivative, quiet, diag, epsfcn, debug)
>     821         return
>     822     self.errmsg = ""
> **--> 824 [self.status, fvec] = self.call(fcn, self.params, functkw)**
>     826 if self.status < 0:
>     827     self.errmsg = f"ERROR: first call to {fcn} failed"
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\external\mpfit\mpfit.py:1337, in mpfit.call(self, fcn, x, functkw, fjac)
>    1335 self.nfev += 1
>    1336 if fjac is None:
> **-> 1337     [status, f] = fcn(x, fjac=fjac, **functkw)**
>    1338     if self.damp > 0:
>    1339         # Apply the damping if requested. This replaces the residuals
>    1340         # with their hyperbolic tangent. Thus residuals larger than
>    1341         # DAMP are essentially clipped.
>    1342         f = np.tanh(f / self.damp)
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\model.py:1333, in BaseModel._errfunc4mpfit(self, p, fjac, x, y, weights)
>    1331 def _errfunc4mpfit(self, p, fjac=None, x=None, y=None, weights=None):
>    1332     if fjac is None:
> **-> 1333         errfunc = self._model_function(p).ravel() - y**
>    1334         if weights is not None:
>    1335             errfunc *= weights.ravel()
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\model.py:1067, in BaseModel._model_function(self, param)
>    1065 self.p0 = param
>    1066 self._fetch_values_from_p0()
> **-> 1067 to_return = self._get_current_data(onlyactive=True, binned=self._binned)**
>    1068 return to_return
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\hyperspy\models\model1d.py:376, in Model1D._get_current_data(self, onlyactive, component_list, binned, ignore_channel_switches)
>     372 if onlyactive:
>     373     component_list = [
>     374         component for component in component_list if component.active
>     375     ]
> --> 376 model_data = self._get_model_data(
>     377     component_list=component_list,
>     378     ignore_channel_switches=ignore_channel_switches,
>     379 )
>     380 if binned is None:
>     381     # use self.axis instead of self.signal.axes_manager[-1]
>     382     # to avoid small overhead (~10 us) which isn't negligeable when
>     383     # __call__ is called repeatably, typically when fitting!
>     384     binned = self.axis.is_binned
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\exspy\models\eelsmodel.py:161, in EELSModel._get_model_data(self, *args, **kwargs)
>     159 for component in component_list:
>     160     if component.convolved:
> --> 161         sum_convolved += component.function(self.convolution_axis)
>     162     else:
>     163         sum_ += component.function(self.axis.axis)
> 
> File ~\AppData\Local\miniconda3\envs\txm\lib\site-packages\exspy\components\eels_cl_edge.py:449, in EELSCLEdge.function(self, E)
>     447 """Returns the number of counts in barns"""
>     448 shift = self.onset_energy.value - self.GOS.onset_energy
> **--> 449 if shift != self.GOS.energy_shift:**
>     450     # Because hspy Events are not executed in any given order,
>     451     # an external function could be in the same event execution list
>     452     # as _integrate_GOS and be executed first. That can potentially
>     453     # cause an error that enforcing _integrate_GOS here prevents. Note
>     454     # that this is suboptimal because _integrate_GOS is computed twice
>     455     # unnecessarily.
>     456     self._integrate_GOS()
>     457 Emax = self.GOS.energy_axis[-1] + self.GOS.energy_shift
> 
> AttributeError: 'GoshGOS' object has no attribute 'energy_shift'

[ericpre]

I can't reproduce the issue: it doesn't raise an error. The error message doesn't make sense and this part of the code hasn't changed recently, so we wouldn't expect a recent regression.
Could it be that you have a broken install? Can you try with a clean install?

It may be worth trying something more streamlined which has been implemented recently: https://hyperspy.org/exspy/user_guide/eels.html#id1

To try to reproduce, here is the code that I adapted your example given above (simply changed the data) :

import exspy
import hyperspy.api as hs

cluster_low_loss = exspy.data.EELS_low_loss()
cluster_high_loss = exspy.data.EELS_MnFe()

s_Mn_L = cluster_high_loss.isig[610.0:700.0].deepcopy()
s_Mn_L.add_elements(('Mn','O'))
model_Mn = s_Mn_L.create_model(low_loss=cluster_low_loss, auto_add_edges=False, auto_background=True,)
model_Mn.components

# hs.set_log_level('ERROR')
model_Mn.fit_component(model_Mn.components.PowerLaw, signal_range=(610.0, 630.0), fit_independent=True, only_current=True)
model_Mn.assign_current_values_to_all(components_list = [model_Mn.components.PowerLaw])
# model_Mn.fit_component(model_Mn.components.PowerLaw, signal_range=(610.0, 630.0), fit_independent=True, only_current=False)


# Create and set the 4 components
import exspy as epy
L3 = epy.components.EELSCLEdge("Mn_L3")
L2 = epy.components.EELSCLEdge("Mn_L2")

L3_white = hs.model.components1D.Gaussian()
L3_white.name = "Mn_L3 line"
L3_white.centre.bmin = 640.0
L3_white.centre.bmax = 645.0
L3_white.centre.value = (L3_white.centre.bmax + L3_white.centre.bmin) / 2 # Initial value of the centre. Must be set to initialise the twinning below.
L3_white.A.bmin = 0
L3_white.sigma.bmax=3

L2_white = hs.model.components1D.Gaussian()
L2_white.name = "Mn_L2 line"
L2_white.centre.bmin = 650.0
L2_white.centre.bmax = 655.0
L2_white.centre.value = (L2_white.centre.bmax + L2_white.centre.bmin) / 2
L2_white.A.bmin = 0
L2_white.sigma.bmax=3

L3.onset_energy.twin = L3_white.centre
L2.onset_energy.twin = L2_white.centre

# Add all four to the model
model_Mn.extend([L3_white, L3, L2_white, L2])
model_Mn.components


# The following sort of emulates smart_fit, but it is more appropriate than smart_fit, 
# because smart_fit doesn't know that it should fit the gaussians between L3 and L2

# We fit "from the left", starting with the lowest energy
model_Mn.reset_signal_range()
model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True,) # get good values
model_Mn.assign_current_values_to_all(components_list = [L3_white]) # apply those good values to all pixels
model_Mn.fit_component(L3_white, signal_range=[L3_white.centre.bmin-3.0, L3_white.centre.bmax+3.0], bounded=True, only_current=False)

[liuchzzyy]

Thanks @ericpre. with the exspy data, it works well on my pc. I am checking my data.