My data looks a bit like a Gaussian (Normal) curve. I fitted it using 1, 2, 3 and 4 gaussians added together where the sigmas, centre positions and heights were variable. It worked well for the 2 and 4 gaussians. I noticed for function with odd number of gaussians added together as 'res', the final fit (based on the x values outputted after cnt # of iterations) height doesn't match the data. i.e if you just make a simple Gaussian function as the data you want to fit and have your guess function as a Gaussian with the same parameters you created the data with, the final x residual/variables will not be the same. The height will be off by a noticeable amount. This only happened with odd numbers though. For even number of gaussians added together in the 'guess' fit, it worked perfectly.
I figured it out! it took a while, i just split the function in half and kept the rest of the form the same res=@(y)eqn(1)(x<b)+eqn(2)(b<=x). took a few tries to get the syntax right. Thanks for the amazing algorithm!!!
I'm fitting with a discontinuous function. My data resembles a normal curve, but is wider on one side so I hope to fit it with 2 halves of a normal curve with different sigmas (spreads) on either side of the centre point. How do I input this into the res function. The residuals are the height, the two sigmas and the centrepoint. Furthermore, will the fact that the centre is what I'm trying to make more accurate a big problem if I'm using it to identify which side of the function (which sigma) I'm on?