User_Info Not Working Correctly in Matlab

[emrdig]

It seems that there is a bug with the call to gpufit within Matlab when including the user_info parameter. Using the included linear_1d model (which utilizes the user_info parameter), I created a simple program in Matlab to model the equation y=x from x=0 to x=10 and called gpufit on the data. This should return the parameters 0 and 1, but results in 4.3467 and 0.8711 instead.

Additionally, if I leave the user_info parameter empty when calling gpufit, it returns the correct values (since it uses x=point_index which would result in x from 0 to 10). This shows that there is a bug with the user_info parameter when calling gpufit from Matlab. A simple script showing this problem is included below.

x=0:10;
params=single([0;1]);
y=single(params(1)+params(2)*x');   % y=x;

tolerance=1e-9;
max_n_iterations=1000;
estimator_id = EstimatorID.LSE;
model_id = ModelID.LINEAR_1D;
init = single([0;1]);   % should result in correct parameters with only 1 iteration

[parameters,states,chi_squares,n_iterations,time]=gpufit(y,[],model_id,init,tolerance,max_n_iterations,[],estimator_id,x);
%[parameters,states,chi_squares,n_iterations,time]=gpufit(y,[],model_id,init,tolerance,max_n_iterations,[],estimator_id,[]);
parameters

[adrianjp88]

The 1D linear model requires the user_info parameter to be of type single. Since the default type in Matlab is double, the expression x = 0:10 in your example defines x as double. Try x = single(0:10).

[emrdig]

Figured it out eventually. It should say this in the Matlab documentation instead of saying "arbitrary type" since the other inputs explicitly say they need to be singles.

[jkfindeisen]

I improved the documentation in the last commit and took your code to create a new example in Matlab for the linear regression with custom x values.

[odewit8]

I had the same issue with the python binding. It worked when I set user_info to numpy.float32. Should change that part of the documentation as well.