Funcfit for multivariable (over 20) equation
imksh2000
Sat, 10/10/2015 - 07:50 pm
Please see attached exp files.
The algorithm is following:
1) Load experimental data wave (not included in the attached file)
2) After recall function setup_display_range() and calculation(), open parameter control macro
3) Setup parameters in order to obtain similar generated graph for the experimental data
// THE FOLLOWING IS FUNCFIT BUT I CANNOT //
4) Save parameters in the panels as initial value of fitting
5) Run funcfit function
For 4) and 5), I should recall all the parameters as coefficient wave and WRITE DOWN ALL THE EQUATION SUCH AS F_uc, F_ctr, and so on.
Is there are clear way to design simple function to get funcfit?
ControlInfo SetVar0
coefwave[0] = V_value
ControlInfo SetVar1
coefwave[1] = V_value
ControlInfo SetVar2
coefwave[2] = V_value
... etc. ...
FuncFit myHugeFitFunct, coefwave, ...
John Weeks
WaveMetrics, Inc.
support@wavemetrics.com
October 12, 2015 at 09:51 am - Permalink
Thank you for your sincere help.
Unfortunately, my question was more focused on how to build myHugeFitFunct.
As you can see in the experiments, I want to utilize CALCULATION() function as myHugeFitFunc.
Is there any nice solution for that?
October 12, 2015 at 02:47 pm - Permalink
DisplayHelpTopic "Fitting to a User-Defined Function"
That topic talks about using the Curve Fit dialog to build and use a user-defined fit function. For your purposes, you will want to edit the Calculation function to make it into a fit function. To learn the proper format, read this topic:
DisplayHelpTopic "User-Defined Fitting Function: Detailed Description"
It's hard to tell from your code what the independent variable is (variables?) or which are constants. The use of global variables for adjustable parameters is discouraged- it makes it hard to debug problems, and easy to make mistakes in setting up to run the function. It is also inefficient for use as a fitting function.
If you want to use Calculation() as the engine for both simulations and for fitting, I would do this:
1) Use the Constant keyword to define numbers that are truly constants, that you will never change. I presume that the global variable that you declare this way:
variable/g atomic_frac_Y = 0.148are actually constants. That one could be defined this way, above the definition of the function:
Constant atomic_frac_Y = 0.1482) Make the adjustable parameters into input parameters. Assuming that the uninitialized globals are actually adjustable parameters, you would do something like
Variable atomic_frac_Zr, atomic_frac_Ovoid, ...
3) It's probably easiest to use the standard fit function format that calls the function once for each X value, and returns one Y value for each. I see
wave Q_x, atomic_positionat the top of Calculation(). Assuming that Q_x is the X value, now the function would look likeVariable Q_x
Variable atomic_frac_Zr, atomic_frac_Ovoid, ...
.... lots of code ....
return <the calculated value>
end
4) Now, to create a fit function you might write this:
Wave pw
Variable Q_x
return Calculation(Q_x, pw[0], pw[1], ... lots more ...)
end
To be sure, I see some things that suggest that perhaps you need to compute all results at the same time. In the detailed description help cited above, you need to read about all-at-once fit functions. They take in an X wave (or waves if it is a multivariate fit) and set the Y values in the Y wave input parameter. They return nothing. All-at-once functions are more challenging to write and debug but generally run faster. Some problems can only be computed using an all-at-once fit function.
John Weeks
WaveMetrics, Inc.
support@wavemetrics.com
October 13, 2015 at 12:39 pm - Permalink