Closed 153957 closed 8 years ago
Thank you very much for your nice bug report. I observed that the Gauss/Landau convolution is not stable if eta < 10 * sigma for the number of fixed number of convolution steps done (100). Thus the convolution steps are now increased depending on the eta / sigma ratio. This should give more stable results; especially when fitting is involved.
The fix is uploaded to PyPI.
When using the langau_pdf with sigma > 10 oscillations appear in the results. In our (HiSPARC) own code we also have an implementation of the Landau PDF and a convolved version: https://github.com/HiSPARC/sapphire/blob/master/sapphire/analysis/landau.py
When I plotted the results for the two convolved Landau/Gauss functions the pyLandau shows oscillations in the result.
Here is the code (requires the script linked above):
Which results in the following, where blue is pyLandau and black the SAPPHiRE implementation.