Hi. > >>> De : Tanguy Yannick [mailto:yannick.tan...@cnes.fr] About the default > >>> value (1e-14), I know the tolerance is case- dependant, but some > >>> times, we don't precisely know the order of magnitude (ex : jacobians > >>> matrices) and it's quite practical to have a default value. > >> > >> Hi, > >> A point of the discussion with Gilles in the MATH-863 (new Quaternion > >> class) is about this magic 1E-14. ;) > >> > >> Consider the following methods: > >> > >> public Quaternion normalize() { > >> final double norm = getNorm(); > >> if (norm >= Precision.SAFE_MIN) { > >> return new Quaternion(q0 / norm, q1 / norm, q2 / norm, q3 / > >> norm); > >> } else { > >> throw new ZeroException(); > >> } > >> } > >> > >> public boolean isUnitQuaternion() { > >> final double norm = getNorm(); > >> if (Precision.equals(norm, 1.0, Precision.EPSILON)) { > >> return true; > >> } else { > >> return false; > >> } > >> } > >> > >> > >> => There are many cases where > >> > >> myQuaternion.normalize().isUnitQuaternion() > >> returns "false". > >> > >> The Precision.EPSILON is too small. That's why we use this "historical" > >> arbitrary tolerance of 1E-14. > >> > >> > >> Do you think that adding a tolerance parameter in this method > >> isUnitQuaternion() (and in its neighboors) is a good idea ? > >> > >> > >> Julien > >> > >I think a tolerance parameter which defaults to 1E-14 in my view be the best > >option. I have nothing against this magic number; I just don't think that it > >should sit in the general class Precision. I think this is also what makes > >Gilles unhappy (TBC). How about a static final double > >Quaternion.DEFAULT_TOL, instead of Precision.DEFAULT_TOL? > >Sébastien > > Ok, I understand your point of view, it make sense to put specific > epsilon/tolerance value in the corresponding classes (and to justify the > value!!). We'll propose this default value in the Quaternion class (see > feature 863) and will use the same mechanism in our other classes.
My point-of-view is a bit more intolerant. ;-) See my previous post (and the forthcoming commit of the "Quaternion" class. Then we can continue this discussion if you don't agree. Regards, Gilles --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
