\(\renewcommand\AA{\unicode{x212B}}\)

Fit Functions In Python

Introduction

Mantid enables Fit function objects to be produced in python. For example

g = Gaussian()

will make g into a Gaussian function with default values and

g = Gaussian(Height=1, Sigma=0.1)

will make g into a Gaussian function with Height set to 1, Sigma set to 0.1 and PeakCentre set to default value.

One can also make function with attributes such as

p = Polynomial(n=3, A0=1, A1=2, A2=4, A3=3)

One can get and set function parameters and attributes, by array operations, for example:

gSigma = g["Sigma"]
g["Height"] = 1.5

One can also get and set function parameters and attributes, by dot operations, for example:

gSigma = g.Sigma
g.Height = 1.5

Composite Functions

Composite functions can be created by means of the plus operation and individual functions can be accessed by array index operations. For example:

spectrum = LinearBackground() + Gaussian(PeakCentre=1) + Gaussian(PeakCentre=2)
peak1 = spectrum[1]
peak1['Sigma'] = 0.123
spectrum[2] = Lorentzian(PeakCentre=2)

which sets Sigma of the second function (first Gaussian) to 0.123 and changes the third function to a Lorentzian.

Similarly product functions can be constructed using the multiplication operator.

p = ExpDecay() * Gaussian(Height=1,Sigma=0.2)

One can get and set the parameters of a composite function, by array operations, but not by dot operations because the parameter name already contains a dot, for example:

f1Sigma = spectrum["f1.Sigma"]
spectrum["f1.Height"] = 1.5

One can add a function by the +- operator or remove be the del function.

spectrum += Lorentzian(PeakCentre=3)
del spectrum[1]

Also available is the len function and iteration over the member functions:

n_peaks = len(spectrum)
for func in spectrum:
   print(func)

The plus and times operators are associative and so may not preserve a composite function within a composite function as such, but replace it with a list of its member functions. Instead you may use:

spectrum = CompositeFunctionWrapper(LinearBackground(), Gaussian(PeakCentre=1), Gaussian(PeakCentre=2))
P = ProductFunctionWrapper(ExpDecay(), Gaussian(Height=1,Sigma=0.2))

Multi-Domain Functions

Multi-Domain functions can be constructed like this:

md_fun = MultiDomainFunction(Gaussian(PeakCentre=1, Sigma=0.1), Gaussian(PeakCentre=1, Sigma=0.2), ..., Global=['Height'])

Setting Ties

The parameters of functions can be tied or fixed like this:

func1.tie(A0=2.0)
func2.tie({'f1.A2': '2*f0.A1', 'f2.A2': '3*f0.A1 + 1'})
func3.fix('A0')
func4.fix('f2.A2')

Both fixes and ties can be removed by untie:

func.untie('f3.Sigma')

To tie all parameters of the same local name in a composite function, one can use TieAll:

func.tieAll('Sigma')

All members of the composite function must have this parameter (in this case Sigma). Similarly with fixing:

spectrum1.fixAll('FWHM')

Also parameters of a function can be fixed with fixAllParameters and unfixed with untieAllParameters.

c.fixAllParameters()
...
c.untieAllParameters()

Setting Constraints

One can set and remove constraints as follows:

g.constrain("Sigma < 2.0, Height > 7.0")
...
g.unconstrain("Sigma")
g.unconstrain("Height")

comp.constrain("f1.Sigma < 2, f0.Height > 7")
...
comp.unconstrain("f1.Sigma")
comp.unconstrain("f0.Height")

One can all constrain a given parameter in all members of a composite function that have this parameter and also remove such constraints.

comp.constrainAll("Sigma < 1.8")
...
comp.unconstrainAll("Sigma")

Evaluation

One can evaluate functions:

p = Polynomial(n=2, A0=0, A1=0.5, A2=0.5)

print p(1)  # expect 1.0
print p(2)  # expect 3.0
print p(3)  # expect 6.0
print p([0,1,2,3])  # expect [ 0. 1. 3. 6.]

ws = CreateWorkspace(DataX=[0,1,2,3,4,5,6,7], DataY=[5,5,5,5,5,5,5])

print p(ws).readY(1) # expect [  0.375   1.875   4.375   7.875  12.375  17.875  24.375]

One can use numpy arrays:

import numpy as np

a = np.array([[0, 1,], [2, 3]])
p = Polynomial(n=4, A0=1, A1=1, A2=1, A3=1, A4=1)
print p(a)
# expect [[   1.    5.]
#         [  31.  121.]]

Also one can put parameters into the function when evaluating.

p = Polynomial(n=2)
print p([0,1,2,3], 0.0, 0.5, 0.5) #expect [ 0. 1. 3. 6.]

This enables one to fit the functions with scipy.optimize.curve_fit.

Errors

The errors assoicated with a given parameter can be accessed using the getError method. getError takes either the parameter name or index as input. For example to get the error on A1 in the above polynomial, the code is:

# Find the parameter error by index
error_A1 = p.getError(1)
# Find the parameter error by name
error_A1 = p.getError('A1')

Plotting

Functions may be plotted by calling the plot method of the function.

This method can be called in any of the following manners:

f.plot(xValues=[0,2,2.5,3,5]) # for these x-values
f.plot(workspace=ws) # for the x-values of workspace ws
f.plot(workspace=ws, workspaceIndex=i)   # for x-values of workspace index i of ws
f.plot(startX=xmin, endX=xmax)  # for 20 x-values between xmin and xmax
f.plot(startX=xmin, endX=xmax, nSteps=10) # for 10 x-values between xmin and xmax
f.plot(workspace=ws, startX=xmin, endX=xmax) # for x-values of ws between xmin & xmax
f.plot(workspace=ws, name='Fred') # for plot & its workspace to be called 'Fred'

If you use xValues, then the list of x values must be in numerical order. This is not checked and if they are not in order the plot may fail to display properly.

If you want to display multiple plots of the same function, then use name to give each plot a unique name. The default value of name is the name of the function.

Category: Concepts