I plotted time series electricity load data in matlab and then used matlab "basic fitting" tools for best fitting. Now i require the values of the best fitted curve, is there a way to extract these values using a code or any other way??

I plotted time series electricity load data in matlab and then used matlab "basic fitting" tools for best fitting of the plotted data (as it represented many electricity load consumers in one plot). Now i require the values of the best fitted curve, is there a way to extract these values using a code or any other way??

[KSSV]

What fitting you have used? Very much you can get extract the values. You have to mention your code so that we can see what you have used.

[zafar khan]

Unfortunately, I do not have any code. What i did is simply opened the figure and used the "Tools" tab to open "basic fitting" tool. From this tool I used different fits including different polynomials and interpolations (obviously i had many figures so each one had a different fit). Now i need the fitted curves values but do not know how to extract them as there are large number of fits that I require values of and manually taking the values using data cursor will take me months to complete it.

Answer 1

[dpb]

Well, the ease of the tool comes at something of a price; it doesn't have the ability to automatically return everything you've done programmatically. What you can do is open the next panel in the UI (the big right arrow at the bottom and it then will let you save each fit in turn to variable struct in the workplace.

Or, you could step back and use the fitting tools in a command window and compute the various fits directly again using the exploratory work you've done to know what it is that need to compute/save.

[zafar khan]

I have tried to save it in workspace but the struct does not hold the real fitted curve values. For polynomial it holds the coefficients depending upon the degree of polynomial but not the curve data points. It gives residual which is not of any use for my work.

Going back a step is a long process which i want to avoid. Is there no other way to retrieve the fitted curve values?

[dpb]

You have to have had them to do the fitting--where did they go? Or, another approach, where did they come from while you were doing the fitting?

Or, yet again, perhaps you had a figure from somewhere else from which to start? In that case, retrieve the '[X|Y]Data' from it.

And, of course, "It gives residual which is not of any use for my work." isn't totally true. If you have the residual, you can evaluate the polynomial at the input points and derive the original data y values by adding the residuals to yPredicted. (You do need the x-values, granted altho if push came to shove you could solve for the x value that gives the value of the residual to find it, even :) if absolutely had to).

ADDENDUM

Or, perhaps having read the previous incorrectly, you mean the computed values from the fit...that's simple, presuming you do have the actual input data (that I maybe mistakenly read the complaint as that you didn't have at hand any longer) just use the coefficients of the curve structure to evaluate the fit at the input points. WHY TMW chose to not save a full set of diagnostics including the predicted values is also beyond my ken in most (if not all?) of these tools but that's "just the way it is", unfortunately.

I've not used the figure tool so not sure; does it save anything for the other models than the coefficients with which to use to evaluate? Well, let's see---yes, a spline returns a spline object which can be used with ppval it seems and the polynomials are in form consistent with polyfit|polyval so it's not that involved it appears...

[zafar khan]

I do have the original data set which was used to get the curve but i do not know how to use the polynomial coefficient to retrieve the values of curve. Seems like I need to study a lot more. I will get back once I have tried something. For now i have used data cursor for some of the important curves but at the end i will definitely need something.

[dpb]

Well, if you use the fitting tool in the figure and save that to the workspace you get something like--

>>
Variables have been created in the current workspace.
>> fit
fit = 
   type: 'polynomial degree 1'
  coeff: [1.5025 57.9005]
>>

Using them is simply

>> yhat=polyval(fit.coeff,x);
>>

where x is the vector of x values used to calculate the fit or any other set of values for which wish to evaluate the polynomial.

If it's a polynomial it doesn't matter what the degree is, polyval uses the length of the input coefficient vector to determine the order without being given it specifically. If you've used another form such as the spline as I previously noted above, it's similar to use the returned structure for it with the appropriate functions to evaluate splines.

Of course, if you have a large number of these to deal with, it would be much simpler and more efficient to simply use polyfit|polyval directly on the data; that could be automated entirely with little effort it would seem and no reason to "touch anything with human hands"...you would need to have made a record of which form you chose for a given set of data from the exploratory work you've already done, of course, in order to use a different model for different cases if that's what's needed, or find a technique that chooses from one of a set of candidate models (such as best R-sq or somesuch similar metric) by comparing between them programmatically.

None of this should be too hard unless the choices are either one of a "veritable plethora" of candidates or the choice(s) you've made are quite subjective in choosing "best" for any given dataset.

[zafar khan]

Thank you very much. I will give it a try. It seems to be the solution i needed. Thanks for the guidance.

[zafar khan]

I have tried it and it worked for me. Thank you very much once again.

[dpb]

Nice to know... :)