T O P I C R E V I E W |
Afrancisco |
Posted - 11/30/2017 : 10:46:07 AM Hi, Suddenly, the functions such as Gauss, Lorentz or any other fitting function stopped initializing values properly, for instance the width is much smaller that it used to be. I installed recently the 2018 version. I worked with it for a couple of days, but it used to intialize parameters satisfactorily, but since I worked with an origin file from a previous version, in the Peak Analysis Dialog it initalizes a much much thinner peak that it should be. I attach an example.
Could the peakpos() function used to initialize parameters be damaged? I also upload a screen caption of the initializing code of the gaussian, but since I can't modify that I don't see how could I have messed up that
The problem may not be in the width, because I think the function used to initialize is a Gaussian, and if the Amplitude is set very low, then the width must be very thin in order to reach the summit of the peak. In fact, the higher the peak the thinner it initializes.
I have added a function wich is the Gaussian Lorentz Cross Product, but assimetric, and I have set to initialize parameters with the function peakpos(), and it also initializes a very very think peak :( So I guess it's a problem of that initializing function, the peakpos()? How did it manage to get screwed? Could it be that I saved a Graph theme as system theme and it somehow affects the parameter initialization?
How can I restore the function peakpos()?
Thanks, Adrian
Origin Ver. and Service Release (Select Help-->About Origin): Origin pro 2018 b9.5.0.193 Operating System: Windows 7 professional |
3 L A T E S T R E P L I E S (Newest First) |
Hideo Fujii |
Posted - 12/01/2017 : 6:04:02 PM Just an extra comment to Easwar's message... since the initial value of the Center is relatively good, though the Width is not so, probably hill-climbing algorithm gets the right place in the parameter space quite easily and quickly by sliding on the "one-way" slope, I guess. (of course, depending on the function. Gaussian should be a good one.) So, probably this is the reason not many people so far didn't pay attention much to this issue. Does it make sense? --Hideo Fujii, OriginLab
|
easwar |
Posted - 12/01/2017 : 2:48:21 PM Hi Adrian,
Thank you for sending in the raw data. We can reproduce the issue. The NLFit dialog initializes the peak properly. But the Peak Analyzer comes up with a bad width as in your picture.
I went back a few versions and I can see at some point in the past the initialization changed to be like this. We need to check further why.
It is likely the Peak Analyzer is NOT using the peak_pos() function but doing its own initialization which gives a bad value when the data is very noisy.
Even though the initialization is bad, if I hit "fit till converge", the fit proceeded fine and converged.
You mentioned in the post the need for a new function "Gaussian Lorentz Cross Product, but asymmetric". If you need help with creating this function, please e-mail us again.
Sincerely,
Easwar OriginLab
|
easwar |
Posted - 11/30/2017 : 12:28:38 PM Hi Adrian,
We roughly digitized the data from your image and the initial values look okay/same in 2017 and 2018:
Can you send your raw data so we can check? Send it to tech@originlab.com and mention this post.
The peak_pos() function should not behave differently from last version to this version. Do you have some default peak analyzer theme that you created which maybe is forcing some bound on the width? Anyway once we have your data, we can check on our side.
Sincerely,
Easwar OriginLab |
|
|