Handling convergence failure

[saugatac]

In RatingCalculator.calculateNewRating, sometimes the loop for convergence:
while (Math.abs(B - A) > CONVERGENCE_TOLERANCE)
fails to terminate. This possibility is also mentioned in the paper. It would be good if there is a maxiConvergenceIterations variable that can be used to control it.

[goochjs]

Hi @saugatac - thanks for the comment. It's quite a while since I worked on this. Do you have examples of where convergence fails? Is the fix something you can add?

[positron]

@saugatac are you sure? This package was developed in 2013 and from the glicko website:

NOTE: As of February 22, 2012 the Glicko-2 iterative procedure ("Step 5") has been corrected and is now stable.

Which matches my understanding of the glicko2 paper as well. He says the old algorithm had convergence issues but the new one (Illinois Algorithm) does not.

[ny83427]

@positron I also found a infinite loop like @saugatac mentioned that convergence fails.

In the debug mode, I just found that B, A always keep same and the abs difference will have no chance to be lower than accepted threshold. Not sure whether we need introduce a max iterate times control parameter for such cases.