[ https://issues.apache.org/jira/browse/MATH-844?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13434999#comment-13434999 ]
Luc Maisonobe commented on MATH-844: ------------------------------------ In this case, the guesser fails because the function in really far from an harmonic function. It is a triangular periodic function with amplitude +/-3 and period 12, and all sample points are taken as integer abscissa, so values all belong to the integer subset {-3, -2, -1, 0, 1, 2, 3}. This is an (interesting) ill-conditioned case. As small integers are represented exactly even as primitive doubles, all computations are exact and one of the intermediate parameters (c2 in the source code) is a perfect 0 (there are no approximation here, the result is exact), and we divide by c2. I would suggest to raise an exception here, and to store this as a junit test for failure mode. > "HarmonicFitter.ParameterGuesser" sometimes fails to return sensible values > --------------------------------------------------------------------------- > > Key: MATH-844 > URL: https://issues.apache.org/jira/browse/MATH-844 > Project: Commons Math > Issue Type: Bug > Affects Versions: 3.0 > Reporter: Gilles > Priority: Minor > Fix For: 3.1 > > Attachments: MATH-844.test.patch > > > The inner class "ParameterGuesser" in "HarmonicFitter" (package > "o.a.c.m.optimization.fitting") fails to compute a usable guess for the > "amplitude" parameter. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira