[
https://issues.apache.org/jira/browse/MATH-487?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12984614#action_12984614
]
Luc Maisonobe commented on MATH-487:
------------------------------------
What I expect is that in many cases, convergence exception appear in iterative
algorithms where the user does have a criterion setting (number of evaluation,
absolute tolerance, relative tolerance, stability check, whatever). We did have
different messages for all these cases, it was too many.
In order to reduce the number of different exceptions, we could wrap all these
ones into convergence exception as a compromise and given the context the user
could work on his settings. Unifying these different errors makes sense because
often in numerical analysis, the various criterion are linked together, so you
may fix a stability check by changing an absolute tolerance for example.
Of course a single Convergence exception is not perfect but its a compromise
solution.
> Deprecate "ConvergenceException" in MATH_2_X and remove it in trunk
> -------------------------------------------------------------------
>
> Key: MATH-487
> URL: https://issues.apache.org/jira/browse/MATH-487
> Project: Commons Math
> Issue Type: Improvement
> Reporter: Gilles
> Assignee: Gilles
> Priority: Minor
> Fix For: 2.2, 3.0
>
>
> The checked "ConvergenceException" should be deprecated.
> An example usage is in class {{ContinuedFraction}} (package {{util}}), at
> line 153:
> {noformat}
> if (scale <= 0) { // Can't scale
> throw new
> ConvergenceException(LocalizedFormats.CONTINUED_FRACTION_INFINITY_DIVERGENCE,
> x);
> }
> {noformat}
> I think that it should be replaced by a more specific (and unchecked)
> exception that reflects the exact low-level problem:
> {noformat}
> if (scale <= 0) { // Can't scale
> throw new NotStrictlypositiveException(scale);
> }
> {noformat}
> A few lines below that, there is:
> {noformat}
> if (infinite) {
> // Scaling failed
> throw new
> ConvergenceException(LocalizedFormats.CONTINUED_FRACTION_INFINITY_DIVERGENCE,
> x);
> }
> {noformat}
> So it seems that it is not necessary to throw an exception at the place where
> the test on "scale" fails; instead we could have:
> {noformat}
> infinite = true;
> if (scale <= 0) { // Can't scale
> break;
> }
> {noformat}
> and let the check on "infinite" throw the exception:
> {noformat}
> if (infinite) {
> // Scaling failed
> throw new
> NotFiniteNumberException(LocalizedFormats.CONTINUED_FRACTION_DIVERGENCE,
> Double.POSITIVE_INFINITY, x);
> }
> {noformat}
> As shown in the above excerpt, we could also replace two {{enum}}:
> * CONTINUED_FRACTION_INFINITY_DIVERGENCE
> * CONTINUED_FRACTION_NAN_DIVERGENCE
> with a single one:
> * CONTINUED_FRACTION_DIVERGENCE
> because the other bit of information (infinity vs NaN) is already given by
> the first parameter of the message.
> What do you think of these changes?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
