How can I change opts.StartPoint for a custom equation at every loop iteration?

Question

pinar 2020-8-11

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/578068-how-can-i-change-opts-startpoint-for-a-custom-equation-at-every-loop-iteration

编辑： Matt J 2021-12-16

Hello,

I have generated a code to fit a Gaussian CDF to the data points of each participant. Since I generated the code by using only one participant's data as a sample, opts.StartPoint was based on this one participant's data points. But the fits were not very good when I did it like this.

So what I want to do is to change opts.StartPoint at each iteration so that the fit on that iteration is based on the start points selected for that participant.

Below is how I have written the start points initially.

```

% Set up fittype and options.

ft = fittype( 'a*(.5*(1+erf((x-m)/(s*sqrt(2)))))', 'independent', 'x', 'dependent', 'y' );

opts = fitoptions( 'Method', 'NonlinearLeastSquares' );

opts.Display = 'Off';

opts.StartPoint = [0.71149061180612 0.231594386708524 0.488897743920167];

opts.MaxFunEvals = 1000;

```

Thank you!

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Uday Pradhan 2020-8-14

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/578068-how-can-i-change-opts-startpoint-for-a-custom-equation-at-every-loop-iteration#answer_479736

在 MATLAB Online 中打开

Hi Pinar,

It is my understanding that you would like to define custom starting points for your curve - fitting model for different participants. To do this, we can create an n by 3 matrix startPts (n being the number of participants) containing the parameters a, m and s in that order (i.e. the i - th row of startPts is the start point for the i - th participant). Then, we can use a loop to create n different fit objects which have different starting points based on the vectors extracted from startPts. I have a code snippet below that you may find useful.

%startPts is the n - by - 3 matrix containing the start points for each of the 'n' participants;
ft = fittype( 'a*(.5*(1+erf((x-m)/(s*sqrt(2)))))', 'independent', 'x', 'dependent', 'y' );
opts = fitoptions( 'Method', 'NonlinearLeastSquares' );
opts.Display = 'Off';
opts.MaxFunEvals = 1000;
s = cell(n,1); %store the n fitobjects for plotting
for i = 1:n
    opts.StartPoint = startPts(i,:);
    s{i} = fit(x,y,ft,opts);
end

3 个评论
显示 1更早的评论隐藏 1更早的评论

Mirjam Studler 2021-11-24

编辑：Mirjam Studler 2021-11-24

Did you manage to find a solution to generate opts.StartPoint automatically in the code? I actually face the same problem.

Matt J 2021-12-16

编辑：Matt J 2021-12-16

Angelavtc's comment moved here:

Dear @Uday Pradhan your answer is very useful, but your code makes the matrix "startPts" a predefined matrix by the user. Is there a way to make this matrix contain the starting point used by the curve fitting tool?

请先登录，再进行评论。