currently, I cannot figure out why the following won't work...
1. I created a column running from -5 to +5 with a step widht of 0.25 2. I calculated the values in column 2 via: Col(B) = 1/2 - 1/Pi*atan(-Col(A)/4) with 4 just one value for a parameter (A) 3. Plotting the data 4. Trying to fit the data with: y=1/2-1/Pi*(atan(-x/A)); and now A should become 2
--> well... A will be refined to be 2 but the x-y-data totally derivates from the input data! In fact, I receive a sigmoidal function that is almost parallel to my input data and shifted toward smaller y-values.
Has anyone any idea what went wrong or how to fix that problem? I want to adapt this regression for real data and not only for simulated ideal values but when such deviations occur even on ideal values...
Thanks a lot! This problem is really bugging me...
Best regards,
Volker
Origin Version (7.5G) / Operating System (Vista Ultimate)
Not sure, but data may be too noisy, or the initial parameter values may be good enough for hill-climbing? Have you tried the simulation to see how good these parameter values are? (Also, make sure that your parameter values works with the angle system (degree or radian) of the function?)
misfitting atan-function
Posted - 07/15/2007 : 08:53:23 AM
