Hi,
I am using this toolbox to fit dynamic gaussian processes and could see that every different run of trainGParx using "minimize.m" was generating a different set of means and standard deviations (a great discrepancy at each run). Then, I updated the folder "gmpl-matlab" with the newest version of GPML available (gpml-matlab-v4.2-2018-06-11) from http://www.gaussianprocess.org/ and not only the results of "minimize.m" are consistent at every run (With regard to the predicted mean and standard deviation) and it was way faster than using "minimizeDE.m".
The only thing that I had to do was to copy and past the "rewrap.m" function from the older GPML directory (the one provided in this repo) and paste it into the newest one.
Therefore, I suggest to update GPML folder with the newest one available.
Hi,
I am using this toolbox to fit dynamic gaussian processes and could see that every different run of trainGParx using "minimize.m" was generating a different set of means and standard deviations (a great discrepancy at each run). Then, I updated the folder "gmpl-matlab" with the newest version of GPML available (gpml-matlab-v4.2-2018-06-11) from http://www.gaussianprocess.org/ and not only the results of "minimize.m" are consistent at every run (With regard to the predicted mean and standard deviation) and it was way faster than using "minimizeDE.m".
The only thing that I had to do was to copy and past the "rewrap.m" function from the older GPML directory (the one provided in this repo) and paste it into the newest one.
Therefore, I suggest to update GPML folder with the newest one available.