When directly fitting a sasmodels model from bumps the best fit theory function is not stored

[pkienzle]

It looks like I didn't implement problem.save() for all sasdata types in the sassmodels bumps wrapper:

sasmodels/sasmodels/bumps_model.py

Lines 277 to 285 in 59f5bae

	def save(self, basename):
	# type: (str) -> None
	"""
	Save the model parameters and data into a file.
	Not Implemented except for sesans fits.
	"""
	if self.data_type == "sesans":
	np.savetxt(basename+".dat", np.array([self._data.x, self.theory()]).T)

That's because the data object for sasmodels is messy:

sasmodels/sasmodels/direct_model.py

Lines 203 to 292 in 59f5bae

	def _interpret_data(self, data: Data, model: KernelModel) -> None:
	# not type: (Data, KernelModel) -> None
	# pylint: disable=attribute-defined-outside-init
	self._data = data
	self._model = model
	# interpret data
	if getattr(data, 'isSesans', False):
	self.data_type = 'sesans'
	elif hasattr(data, 'qx_data'):
	self.data_type = 'Iqxy'
	elif getattr(data, 'oriented', False):
	self.data_type = 'Iq-oriented'
	else:
	self.data_type = 'Iq'
	if self.data_type == 'sesans':
	res = _make_sesans_transform(data)
	index = slice(None, None)
	if data.y is not None:
	Iq, dIq = data.y, data.dy
	else:
	Iq, dIq = None, None
	elif self.data_type == 'Iqxy':
	#if not model.info.parameters.has_2d:
	# raise ValueError("not 2D without orientation or magnetic parameters")
	q = np.sqrt(data.qx_data**2 + data.qy_data**2)
	qmin = getattr(data, 'qmin', 1e-16)
	qmax = getattr(data, 'qmax', np.inf)
	accuracy = getattr(data, 'accuracy', 'Low')
	index = (data.mask == 0) & (q >= qmin) & (q <= qmax)
	if data.data is not None:
	index &= ~np.isnan(data.data)
	Iq = data.data[index]
	dIq = data.err_data[index]
	else:
	Iq, dIq = None, None
	res = resolution2d.Pinhole2D(data=data, index=index,
	nsigma=3.0, accuracy=accuracy)
	elif self.data_type == 'Iq':
	index = (data.x >= data.qmin) & (data.x <= data.qmax)
	mask = getattr(data, 'mask', None)
	if mask is not None:
	index &= (mask == 0)
	if data.y is not None:
	index &= ~np.isnan(data.y)
	Iq = data.y[index]
	dIq = data.dy[index]
	else:
	Iq, dIq = None, None
	if getattr(data, 'dx', None) is not None:
	q, dq = data.x[index], data.dx[index]
	if (dq > 0).any():
	res = resolution.Pinhole1D(q, dq)
	else:
	res = resolution.Perfect1D(q)
	elif (getattr(data, 'dxl', None) is not None
	or getattr(data, 'dxw', None) is not None):
	res = resolution.Slit1D(
	data.x[index],
	q_length=None if data.dxl is None else data.dxl[index],
	q_width=None if data.dxw is None else data.dxw[index])
	else:
	res = resolution.Perfect1D(data.x[index])
	elif self.data_type == 'Iq-oriented':
	index = (data.x >= data.qmin) & (data.x <= data.qmax)
	if data.y is not None:
	index &= ~np.isnan(data.y)
	Iq = data.y[index]
	dIq = data.dy[index]
	else:
	Iq, dIq = None, None
	if (getattr(data, 'dxl', None) is None
	or getattr(data, 'dxw', None) is None):
	raise ValueError("oriented sample with 1D data needs slit resolution")
	res = resolution2d.Slit2D(
	data.x[index],
	qx_width=data.dxw[index],
	qy_width=data.dxl[index])
	else:
	raise ValueError("Unknown data type") # never gets here
	# Remember function inputs so we can delay loading the function and
	# so we can save/restore state
	self._kernel = None
	self.Iq, self.dIq, self.index = Iq, dIq, index
	self.resolution = res
	self.results = None # type: Optional[Callable[[], OrderedDict]]

To add simple 1D sans data I think we could rewrite the save as follows (untested):

def save(self, basename):
    # type: (str) -> None
    """
    Save the model parameters and data into a file.

    Not Implemented except for sesans fits.
    """
    if self.data_type == "sesans":
        np.savetxt(basename+".dat", np.array([self._data.x, self.theory()]).T)
    elif self.data_type == "Iq":
        data, index = self._data.data, self._data.index
        x, y, dy, dx = data.x[index], self.theory(), data.dy[index], data.dx[index]
        np.savetxt(basename+".dat", np.array([x, y, dy, dx]).T)

For now you can monkey-patch this into sasmodels at the start of your script:

from sasmodels.bumps_model import Experiment
Experiment.save = save

[pkienzle]

… and extend this for the other data types return by sasdata.

Particularly messy is the adhoc masking I do within the _interpret_data method, which means the Iq returned by the theory calculation doesn't align with the measured q or (qx,qy) values.

Related: the complexity required for save already lives in the sasmodels plotting routines.

We could have a method to set set up the data structures needed for save/plot in one function so that they only need to render the data to a file in save or to a graph in plot.