def makeRecipe(datname):
"""Create a fitting recipe for ellipsoidal SAS data."""
## The Profile
# This will be used to store the observed and calculated I(Q) data.
profile = Profile()
# Load data and add it to the Profile. We use a SASParser to load the data
# properly and pass the metadata along.
parser = SASParser()
parser.parseFile(datname)
profile.loadParsedData(parser)
## The ProfileGenerator
# The SASGenerator is for configuring and calculating a SAS profile. We use
# a sans model to configure and serve as the calculation engine of the
# generator. This allows us to use the full sans model creation
# capabilities, and tie this into SrFit when we want to fit a model to
# data. The documentation for the various sans models can be found at
# http://danse.chem.utk.edu/sansview.html.
from sans.models.EllipsoidModel import EllipsoidModel
model = EllipsoidModel()
generator = SASGenerator("generator", model)
## The FitContribution
# Here we associate the Profile and ProfileGenerator, as has been done
# before.
contribution = FitContribution("ellipsoid")
contribution.addProfileGenerator(generator)
contribution.setProfile(profile, xname="q")
# We want to fit the log of the signal to the log of the data so that the
# higher-Q information remains significant. There are no I(Q) uncertainty
# values with the data, so we do not need to worry about the effect this
# will have on the estimated parameter uncertainties.
contribution.setResidualEquation("log(eq) - log(y)")
## Make the FitRecipe and add the FitContribution.
recipe = FitRecipe()
recipe.addContribution(contribution)
## Configure the fit variables
# The SASGenerator uses the parameters from the params and dispersion
# attribues of the model. These vary from model to model, but are adopted
# as SrFit Parameters within the generator. Whereas the dispersion
# parameters are accessible as, e.g. "radius.width", within the
# SASGenerator these are named like "radius_width".
#
# We want to fit the scale factor, radii and background factors.
recipe.addVar(generator.scale, 1)
recipe.addVar(generator.radius_a, 50)
recipe.addVar(generator.radius_b, 500)
recipe.addVar(generator.background, 0)
# Give the recipe away so it can be used!
return recipe
def makeRecipe(datname):
"""Create a fitting recipe for ellipsoidal SAS data."""
## The Profile
# This will be used to store the observed and calculated I(Q) data.
profile = Profile()
# Load data and add it to the Profile. We use a SASParser to load the data
# properly and pass the metadata along.
parser = SASParser()
parser.parseFile(datname)
profile.loadParsedData(parser)
## The ProfileGenerator
# The SASGenerator is for configuring and calculating a SAS profile. We use
# a sans model to configure and serve as the calculation engine of the
# generator. This allows us to use the full sans model creation
# capabilities, and tie this into SrFit when we want to fit a model to
# data. The documentation for the various sans models can be found at
# http://danse.chem.utk.edu/sansview.html.
from sans.models.EllipsoidModel import EllipsoidModel
model = EllipsoidModel()
generator = SASGenerator("generator", model)
## The FitContribution
# Here we associate the Profile and ProfileGenerator, as has been done
# before.
contribution = FitContribution("ellipsoid")
contribution.addProfileGenerator(generator)
contribution.setProfile(profile, xname = "q")
# We want to fit the log of the signal to the log of the data so that the
# higher-Q information remains significant. There are no I(Q) uncertainty
# values with the data, so we do not need to worry about the effect this
# will have on the estimated parameter uncertainties.
contribution.setResidualEquation("log(eq) - log(y)")
## Make the FitRecipe and add the FitContribution.
recipe = FitRecipe()
recipe.addContribution(contribution)
## Configure the fit variables
# The SASGenerator uses the parameters from the params and dispersion
# attribues of the model. These vary from model to model, but are adopted
# as SrFit Parameters within the generator. Whereas the dispersion
# parameters are accessible as, e.g. "radius.width", within the
# SASGenerator these are named like "radius_width".
#
# We want to fit the scale factor, radii and background factors.
recipe.addVar(generator.scale, 1)
recipe.addVar(generator.radius_a, 50)
recipe.addVar(generator.radius_b, 500)
recipe.addVar(generator.background, 0)
# Give the recipe away so it can be used!
return recipe
def make_contribution(config: ConConfig) -> FitContribution:
"""
Make a FitContribution according to the ConConfig.
Parameters
----------
config : ConConfig
The configuration instance for the FitContribution.
Returns
-------
contribution : FitContribution
The FitContribution built from ConConfig.
"""
contribution = FitContribution(config.name)
fit_range = config.fit_range
profile = make_profile(config.data_file, fit_range)
contribution.setProfile(profile, xname="r")
for phase in config.phases:
generator = make_generator(phase)
generator.qdamp.value = config.qparams[0]
generator.qbroad.value = config.qparams[1]
contribution.addProfileGenerator(generator)
for base_line in config.base_lines:
contribution.addProfileGenerator(base_line)
for function in config.functions:
name = function.name
func_type = function.func_type
argnames = function.argnames
contribution.registerFunction(func_type, name, argnames)
contribution.setEquation(config.eq)
contribution.setResidualEquation(config.res_eq)
return contribution
def makeRecipe(ciffile, xdatname, ndatname):
"""Create a fitting recipe for crystalline PDF data."""
## The Profiles
# We need a profile for each data set. This means that we will need two
# FitContributions as well.
xprofile = Profile()
nprofile = Profile()
# Load data and add it to the proper Profile.
parser = PDFParser()
parser.parseFile(xdatname)
xprofile.loadParsedData(parser)
xprofile.setCalculationRange(xmax = 20)
parser = PDFParser()
parser.parseFile(ndatname)
nprofile.loadParsedData(parser)
nprofile.setCalculationRange(xmax = 20)
## The ProfileGenerators
# We need one of these for the x-ray data.
xgenerator = PDFGenerator("G")
stru = loadCrystal(ciffile)
xgenerator.setStructure(stru)
# And we need one for the neutron data. We want to refine the same
# structure object in each PDFGenerator. This would suggest that we add the
# same Crystal to each. However, if we do that then we will have two
# Parameters for each Crystal data member (two Parameters for the "a"
# lattice parameter, etc.), held in different ObjCrystCrystalParSets, each
# managed by its own PDFGenerator. Thus, changes made to the Crystal
# through one PDFGenerator will not be known to the other PDFGenerator
# since their ObjCrystCrystalParSets don't know about each other. The
# solution is to share ObjCrystCrystalParSets rather than Crystals. This
# way there is only one Parameter for each Crystal data member. (An
# alternative to this is to constrain each structure Parameter to be varied
# to the same variable. The present approach is easier and less error
# prone.)
#
# Tell the neutron PDFGenerator to use the phase from the x-ray
# PDFGenerator.
ngenerator = PDFGenerator("G")
ngenerator.setPhase(xgenerator.phase)
## The FitContributions
# We associate the x-ray PDFGenerator and Profile in one FitContribution...
xcontribution = FitContribution("xnickel")
xcontribution.addProfileGenerator(xgenerator)
xcontribution.setProfile(xprofile, xname = "r")
# and the neutron objects in another.
ncontribution = FitContribution("nnickel")
ncontribution.addProfileGenerator(ngenerator)
ncontribution.setProfile(nprofile, xname = "r")
# This example is different than the previous ones in that we are composing
# a residual function from other residuals (one for the x-ray contribution
# and one for the neutron contribution). The relative magnitude of these
# residuals effectively determines the influence of each contribution over
# the fit. This is a problem in this case because the x-ray data has
# uncertainty values associated with it (on the order of 1e-4), and the
# chi^2 residual is proportional to 1 / uncertainty**2. The neutron has no
# uncertainty, so it's chi^2 is proportional to 1. Thus, my optimizing
# chi^2 we would give the neutron data practically no weight in the fit. To
# get around this, we will optimize a different metric.
#
# The contribution's residual can be either chi^2, Rw^2, or custom crafted.
# In this case, we should minimize Rw^2 of each contribution so that each
# one can contribute roughly equally to the fit.
xcontribution.setResidualEquation("resv")
ncontribution.setResidualEquation("resv")
# Make the FitRecipe and add the FitContributions.
recipe = FitRecipe()
recipe.addContribution(xcontribution)
recipe.addContribution(ncontribution)
# Now we vary and constrain Parameters as before.
recipe.addVar(xgenerator.scale, 1, "xscale")
recipe.addVar(ngenerator.scale, 1, "nscale")
recipe.addVar(xgenerator.qdamp, 0.01, "xqdamp")
recipe.addVar(ngenerator.qdamp, 0.01, "nqdamp")
# delta2 is a non-structual material propery. Thus, we constrain together
# delta2 Parameter from each PDFGenerator.
delta2 = recipe.newVar("delta2", 2)
recipe.constrain(xgenerator.delta2, delta2)
recipe.constrain(ngenerator.delta2, delta2)
# We only need to constrain phase properties once since there is a single
# ObjCrystCrystalParSet for the Crystal.
phase = xgenerator.phase
for par in phase.sgpars:
recipe.addVar(par)
recipe.B11_0 = 0.1
# Give the recipe away so it can be used!
return recipe
def makeRecipe(ciffile, grdata, iqdata):
"""Make complex-modeling recipe where I(q) and G(r) are fit
simultaneously.
The fit I(q) is fed into the calculation of G(r), which provides feedback
for the fit parameters of both.
"""
# Create a PDF contribution as before
pdfprofile = Profile()
pdfparser = PDFParser()
pdfparser.parseFile(grdata)
pdfprofile.loadParsedData(pdfparser)
pdfprofile.setCalculationRange(xmin = 0.1, xmax = 20)
pdfcontribution = FitContribution("pdf")
pdfcontribution.setProfile(pdfprofile, xname = "r")
pdfgenerator = PDFGenerator("G")
pdfgenerator.setQmax(30.0)
stru = loadCrystal(ciffile)
pdfgenerator.setStructure(stru)
pdfcontribution.addProfileGenerator(pdfgenerator)
pdfcontribution.setResidualEquation("resv")
# Create a SAS contribution as well. We assume the nanoparticle is roughly
# elliptical.
sasprofile = Profile()
sasparser = SASParser()
sasparser.parseFile(iqdata)
sasprofile.loadParsedData(sasparser)
if all(sasprofile.dy == 0):
sasprofile.dy[:] = 1
sascontribution = FitContribution("sas")
sascontribution.setProfile(sasprofile)
from sas.models.EllipsoidModel import EllipsoidModel
model = EllipsoidModel()
sasgenerator = SASGenerator("generator", model)
sascontribution.addProfileGenerator(sasgenerator)
sascontribution.setResidualEquation("resv")
# Now we set up a characteristic function calculator that depends on the
# sas model.
cfcalculator = SASCF("f", model)
# Register the calculator with the pdf contribution and define the fitting
# equation.
pdfcontribution.registerCalculator(cfcalculator)
# The PDF for a nanoscale crystalline is approximated by
# Gnano = f * Gcryst
pdfcontribution.setEquation("f * G")
# Moving on
recipe = FitRecipe()
recipe.addContribution(pdfcontribution)
recipe.addContribution(sascontribution)
# PDF
phase = pdfgenerator.phase
for par in phase.sgpars:
recipe.addVar(par)
recipe.addVar(pdfgenerator.scale, 1)
recipe.addVar(pdfgenerator.delta2, 0)
# SAS
recipe.addVar(sasgenerator.scale, 1, name = "iqscale")
recipe.addVar(sasgenerator.radius_a, 10)
recipe.addVar(sasgenerator.radius_b, 10)
# Even though the cfcalculator and sasgenerator depend on the same sas
# model, we must still constrain the cfcalculator Parameters so that it is
# informed of changes in the refined parameters.
recipe.constrain(cfcalculator.radius_a, "radius_a")
recipe.constrain(cfcalculator.radius_b, "radius_b")
return recipe
def makeRecipe(ciffile, grdata, iqdata):
"""Make complex-modeling recipe where I(q) and G(r) are fit
simultaneously.
The fit I(q) is fed into the calculation of G(r), which provides feedback
for the fit parameters of both.
"""
# Create a PDF contribution as before
pdfprofile = Profile()
pdfparser = PDFParser()
pdfparser.parseFile(grdata)
pdfprofile.loadParsedData(pdfparser)
pdfprofile.setCalculationRange(xmin = 0.1, xmax = 20)
pdfcontribution = FitContribution("pdf")
pdfcontribution.setProfile(pdfprofile, xname = "r")
pdfgenerator = PDFGenerator("G")
pdfgenerator.setQmax(30.0)
stru = CreateCrystalFromCIF(file(ciffile))
pdfgenerator.setStructure(stru)
pdfcontribution.addProfileGenerator(pdfgenerator)
pdfcontribution.setResidualEquation("resv")
# Create a SAS contribution as well. We assume the nanoparticle is roughly
# elliptical.
sasprofile = Profile()
sasparser = SASParser()
sasparser.parseFile(iqdata)
sasprofile.loadParsedData(sasparser)
sascontribution = FitContribution("sas")
sascontribution.setProfile(sasprofile)
from sans.models.EllipsoidModel import EllipsoidModel
model = EllipsoidModel()
sasgenerator = SASGenerator("generator", model)
sascontribution.addProfileGenerator(sasgenerator)
sascontribution.setResidualEquation("resv")
# Now we set up a characteristic function calculator that depends on the
# sas model.
cfcalculator = SASCF("f", model)
# Register the calculator with the pdf contribution and define the fitting
# equation.
pdfcontribution.registerCalculator(cfcalculator)
# The PDF for a nanoscale crystalline is approximated by
# Gnano = f * Gcryst
pdfcontribution.setEquation("f * G")
# Moving on
recipe = FitRecipe()
recipe.addContribution(pdfcontribution)
recipe.addContribution(sascontribution)
# PDF
phase = pdfgenerator.phase
for par in phase.sgpars:
recipe.addVar(par)
recipe.addVar(pdfgenerator.scale, 1)
recipe.addVar(pdfgenerator.delta2, 0)
# SAS
recipe.addVar(sasgenerator.scale, 1, name = "iqscale")
recipe.addVar(sasgenerator.radius_a, 10)
recipe.addVar(sasgenerator.radius_b, 10)
# Even though the cfcalculator and sasgenerator depend on the same sas
# model, we must still constrain the cfcalculator Parameters so that it is
# informed of changes in the refined parameters.
recipe.constrain(cfcalculator.radius_a, "radius_a")
recipe.constrain(cfcalculator.radius_b, "radius_b")
return recipe
def makeRecipe(ciffile, xdatname, ndatname):
"""Create a fitting recipe for crystalline PDF data."""
## The Profiles
# We need a profile for each data set. This means that we will need two
# FitContributions as well.
xprofile = Profile()
nprofile = Profile()
# Load data and add it to the proper Profile.
parser = PDFParser()
parser.parseFile(xdatname)
xprofile.loadParsedData(parser)
xprofile.setCalculationRange(xmax = 20)
parser = PDFParser()
parser.parseFile(ndatname)
nprofile.loadParsedData(parser)
nprofile.setCalculationRange(xmax = 20)
## The ProfileGenerators
# We need one of these for the x-ray data.
xgenerator = PDFGenerator("G")
stru = CreateCrystalFromCIF(file(ciffile))
xgenerator.setStructure(stru)
# And we need one for the neutron data. We want to refine the same
# structure object in each PDFGenerator. This would suggest that we add the
# same Crystal to each. However, if we do that then we will have two
# Parameters for each Crystal data member (two Parameters for the "a"
# lattice parameter, etc.), held in different ObjCrystCrystalParSets, each
# managed by its own PDFGenerator. Thus, changes made to the Crystal
# through one PDFGenerator will not be known to the other PDFGenerator
# since their ObjCrystCrystalParSets don't know about each other. The
# solution is to share ObjCrystCrystalParSets rather than Crystals. This
# way there is only one Parameter for each Crystal data member. (An
# alternative to this is to constrain each structure Parameter to be varied
# to the same variable. The present approach is easier and less error
# prone.)
#
# Tell the neutron PDFGenerator to use the phase from the x-ray
# PDFGenerator.
ngenerator = PDFGenerator("G")
ngenerator.setPhase(xgenerator.phase)
## The FitContributions
# We associate the x-ray PDFGenerator and Profile in one FitContribution...
xcontribution = FitContribution("xnickel")
xcontribution.addProfileGenerator(xgenerator)
xcontribution.setProfile(xprofile, xname = "r")
# and the neutron objects in another.
ncontribution = FitContribution("nnickel")
ncontribution.addProfileGenerator(ngenerator)
ncontribution.setProfile(nprofile, xname = "r")
# This example is different than the previous ones in that we are composing
# a residual function from other residuals (one for the x-ray contribution
# and one for the neutron contribution). The relative magnitude of these
# residuals effectively determines the influence of each contribution over
# the fit. This is a problem in this case because the x-ray data has
# uncertainty values associated with it (on the order of 1e-4), and the
# chi^2 residual is proportional to 1 / uncertainty**2. The neutron has no
# uncertainty, so it's chi^2 is proportional to 1. Thus, my optimizing
# chi^2 we would give the neutron data practically no weight in the fit. To
# get around this, we will optimize a different metric.
#
# The contribution's residual can be either chi^2, Rw^2, or custom crafted.
# In this case, we should minimize Rw^2 of each contribution so that each
# one can contribute roughly equally to the fit.
xcontribution.setResidualEquation("resv")
ncontribution.setResidualEquation("resv")
# Make the FitRecipe and add the FitContributions.
recipe = FitRecipe()
recipe.addContribution(xcontribution)
recipe.addContribution(ncontribution)
# Now we vary and constrain Parameters as before.
recipe.addVar(xgenerator.scale, 1, "xscale")
recipe.addVar(ngenerator.scale, 1, "nscale")
recipe.addVar(xgenerator.qdamp, 0.01, "xqdamp")
recipe.addVar(ngenerator.qdamp, 0.01, "nqdamp")
# delta2 is a non-structual material propery. Thus, we constrain together
# delta2 Parameter from each PDFGenerator.
delta2 = recipe.newVar("delta2", 2)
recipe.constrain(xgenerator.delta2, delta2)
recipe.constrain(ngenerator.delta2, delta2)
# We only need to constrain phase properties once since there is a single
# ObjCrystCrystalParSet for the Crystal.
phase = xgenerator.phase
for par in phase.sgpars:
recipe.addVar(par)
recipe.B11_0 = 0.1
# Give the recipe away so it can be used!
return recipe
