Le sam. 23 déc. 2023 à 23:50, sebb <seb...@gmail.com> a écrit : > > On Sat, 23 Dec 2023 at 15:44, Gilles Sadowski <gillese...@gmail.com> wrote: > > > > Hello. > > > > Le sam. 23 déc. 2023 à 16:30, sebb <seb...@gmail.com> a écrit : > > > > > > The test MultiStartMultivariateOptimizerTest.testCircleFitting [1] > > > seems to be flaky. > > > > > > It often fails with values below the expected minimum of 750. > > > > > > Does this indicate a real failure, or can the minimum be safely reduced? > > > > I think so (if the right answer is found in fewer iterations, all the > > better). > > I wondered about that, but then why is there a lower limit at all? > > The history shows that the minimum was previously lower, but there are > suggestions that the counts may have been inaccurate. > At one point the limits were 590 < n < 610 which would definitely not work > now!
Testing a lower limit predates the API refactoring (that occurred many years ago). I vaguely recall (?) that the lower limit had to be increased because the previous implementation may have indeed not counted some calls to the evaluation function. Regards, Gilles > > > > > > Sebb > > > [1] > > > https://github.com/apache/commons-math/blob/b0cefc56c67bd41ec9f14311f058d319b05c0312/commons-math-legacy/src/test/java/org/apache/commons/math4/legacy/optim/nonlinear/scalar/MultiStartMultivariateOptimizerTest.java#L75-L76 > > > --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org