Peak fitting algorithms

[Nick Draper]

Need to get a list of 8 algorithms from the steering committee.

Also the structure should allow for user defined algorithms.

1D peak fitting is fine initially.

We already bind to GSL which has some peak fitting functionality. We should be able to wrap up some of that into one or more algorithms to do this.

The steering committee members are: Perring, TG (Toby) Radaelli, PG (Paolo) Soper, AK (Alan) Pratt, FL (Francis) Langridge, S (Sean) Ramirez-Cuesta, AJ (Timmy) -- wasn't present at the meeting Charlton, TR (Timothy) Hughes, AV (Arwel)

[Anders Markvardsen]

(In [1184]) Added GSL based least squares algorithm to fit spectrum to a Gaussian. Refs #139.

[Anders Markvardsen]

(In [1185]) Debug versions of GSL libraries added. Refs #139.

[Matt Clarke]

(In [1189]) Adjusted GaussLeastSquaresFit to build on Linux. Refs #139

[Nick Draper]

(In [1192]) re #139

Corrected debug GSL lib files so they allowed runtime loading of Algorithms dll

[Anders Markvardsen]

(In [1219]) Output summary from Gaussian fit to (output) properties. Refs #139.

[Anders Markvardsen]

(In [1221]) Changed some friend method declarations to standard function declarations to better cross-platform compiling.... Refs #139.

[Anders Markvardsen]

(In [1225]) Fixed (or at least I think I have) a recurring painful linking problem with gsl under windows. In debug mode changed dependency from gsl.dll to gsl_d.dll. Refs #139.

[Anders Markvardsen]

(In [1228]) Modified gauss LS alg. so that initial values for the input parameters are required (otherwise assumed to be zero for now). Refs #139.

[Anders Markvardsen]

(In [1233]) Added algorithm for back-to-back exponential peak shape function. Note, currently the added fitting appears to give a better fit compared to using QTIPLOT. It is currently not understood why there is this discrepancy. Refs #139.

[Anders Markvardsen]

(In [1235]) Refs #139.

[Anders Markvardsen]

(In [1243]) Moved FitData structure into .cpp files. This fixed issue with including multiple fitting algorithm header files. Refs #139.

[Russell Taylor]

(In [1244]) Fix conflicting output workspace names in tests. Re #139.

[Anders Markvardsen]

(In [1329]) Changes to remove some Doxygen warnings. Refs #139.

[Anders Markvardsen]

(In [1334]) Added addition test to GaussLeastSquaresFit and code to take mid-point values for histogram data. Refs #139.

[Anders Markvardsen]

(In [1556]) Renaming of curve fitting algorithms as a consequence of these now living in separate CurveFitting project. Refs #139.

[Anders Markvardsen]

(In [1570]) Changed the names of the fitting parameters for the Gaussian fitting function: xc->peakCentre, A->height, w->sigma, y0->bg0 and brought the documentation for this class up-to-date in the .h and .cpp files. Refs #139.

[Anders Markvardsen]

(In [1573]) GSL may accept a gaussian fit with a negative sigma. For now deal with this by changing the status from 'success' to 'failure: sigma parameter fitted to a value <= 0.0' when this happen. Refs #139.

[Anders Markvardsen]

(In [1574]) Modified Gaussian fitting so that 1/sigma^{2 is least squared fitted instead of sigma directly. This will hopefully avoid the pathological case of GSL reporting a successful fit with a negative sigma... Refs #139.}

[Anders Markvardsen]

(In [1746]) Removed unnecessary reference to background fitting parameter in gauss_df. Refs #139.

[Nick Draper]

moved to unassigned.

The majority of curve fits people want seems to be covered by Gaussian at the moment. We will add other as necessary

[Stuart Campbell]

This ticket has been transferred to github issue ​987