Thanks, Luc. I am fine with the change. You're right that in this (and lots if other cases), we should be either looking at relative error or setting tolerances by element.
On 5/6/09, luc.maison...@free.fr <luc.maison...@free.fr> wrote: > The first element in the reference table has a large magnitude (3.48e6). The > error tolerance was 1.0e-8 which was 14 digits smaller. The change from > RealMatrix to RealVector scheduled the numeric operations differently, > despite the same values should theoretically be computed (the matrix used > was a 1x1 matrix from which the trace was extracted, now we compute a > vectorial dot product which leads to a double). > After the change, the resulting value was about -3482258.63459583 (last > significant digit changed). This seemed acceptable to me. A better change > would probably have been to replace an absolute threshold by a relative > threshold. > > Do you want me to revert the change ? > > Luc > > ----- "Phil Steitz" <phil.ste...@gmail.com> a écrit : > >> > >> > /** >> > >> > Modified: >> commons/proper/math/trunk/src/test/org/apache/commons/math/stat/regression/OLSMultipleLinearRegressionTest.java >> > URL: >> http://svn.apache.org/viewvc/commons/proper/math/trunk/src/test/org/apache/commons/math/stat/regression/OLSMultipleLinearRegressionTest.java?rev=772114&r1=772113&r2=772114&view=diff >> > >> ============================================================================== >> > --- >> commons/proper/math/trunk/src/test/org/apache/commons/math/stat/regression/OLSMultipleLinearRegressionTest.java >> (original) >> > +++ >> commons/proper/math/trunk/src/test/org/apache/commons/math/stat/regression/OLSMultipleLinearRegressionTest.java >> Wed May 6 09:40:13 2009 >> > @@ -139,7 +139,7 @@ >> > new double[]{-3482258.63459582, 15.0618722713733, >> > -0.358191792925910E-01,-2.02022980381683, >> > -1.03322686717359,-0.511041056535807E-01, >> > - 1829.15146461355}, 1E-8); // >> > + 1829.15146461355}, 2E-8); // >> > >> What caused this? >> >> Phil >> >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org >> For additional commands, e-mail: dev-h...@commons.apache.org > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org > For additional commands, e-mail: dev-h...@commons.apache.org > > --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
