You will get the best possible results using the regression fit routines if you adjust your input data arrays so that they contain numbers that are "small". These routines suffer from numeric errors internally when processing large numbers, like timestamps.
I do this to the data before passing it to the regression function by subtracting each input array's minimum value from all array elements. You should do this with both the X and Y input arrays (unless you know the values of the array data will always be close to zero). This must be done if one of the arrays contains timestamps. Then after the fit is calculated, you add the offset back in to the output arrays appropriately. I have recommended to NI tech support several times that NI should enhance the regression functions to automatically do this. They have refused to take my advice. ===================================================== when making a polynomial fit on a data set with n points it should always be possible to find a n-1 degree polynomial expression that gives a mean squared error equal to 0 (well, rounding errors make this an 'almost' 0) When using the SVD algoritm in the General Plynomial Fit.vi this isn't always the case. Is this due to the qualities of the SVD algorithm ? E.g. the Givens algorithm seems more stable. My math is a little rusty so
