float division by zero

[youknowone]

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "trueskill/__init__.py", line 416, in transform_ratings
    return g().transform_ratings(rating_groups, ranks, min_delta)
  File "trueskill/__init__.py", line 341, in transform_ratings
    self.run_schedule(*layers, min_delta=min_delta)
  File "trueskill/__init__.py", line 288, in run_schedule
    teamdiff_layer[0].up(0)
  File "trueskill/factorgraph.py", line 140, in up
    return self.update(self.terms[index], vals, msgs, coeffs)
  File "trueskill/factorgraph.py", line 145, in update
    pi = 1. / sum(coeffs[x] ** 2 / divs[x].pi for x in xrange(size))
  File "trueskill/factorgraph.py", line 145, in <genexpr>
    pi = 1. / sum(coeffs[x] ** 2 / divs[x].pi for x in xrange(size))
ZeroDivisionError: float division by zero

[sublee]

Please include the input ratings that drives the error.

[konikos]

Same for me for the following datasets:

team1 = (Rating(mu=42.234, sigma=3.728), Rating(mu=43.290, sigma=3.842))
team2 = (Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500), Rating(mu=16.667, sigma=0.500))
transform_ratings([team1, team2], ranks=[6, 5])

and

team1 = (Rating(mu=25.000, sigma=0.500), Rating(mu=25.000, sigma=0.500), Rating(mu=25.000, sigma=0.500), Rating(mu=25.000, sigma=0.500), Rating(mu=33.333, sigma=0.500), Rating(mu=33.333, sigma=0.500), Rating(mu=33.333, sigma=0.500), Rating(mu=33.333, sigma=0.500), Rating(mu=41.667, sigma=0.500), Rating(mu=41.667, sigma=0.500), Rating(mu=41.667, sigma=0.500), Rating(mu=41.667, sigma=0.500))
team2 = (Rating(mu=42.234, sigma=3.728), Rating(mu=43.291, sigma=3.842))
transform_ratings([team1, team2], ranks=[0,28])

Using python 2.7 and trueskill 0.1.3 from pypi results to

Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File "/tmp/venv/local/lib/python2.7/site-packages/trueskill/__init__.py", line 414, in transform_ratings
    return g().transform_ratings(rating_groups, ranks, min_delta)
  File "/tmp/venv/local/lib/python2.7/site-packages/trueskill/__init__.py", line 339, in transform_ratings
    self.run_schedule(*layers, min_delta=min_delta)
  File "/tmp/venv/local/lib/python2.7/site-packages/trueskill/__init__.py", line 286, in run_schedule
    teamdiff_layer[0].up(0)
  File "/tmp/venv/local/lib/python2.7/site-packages/trueskill/factorgraph.py", line 140, in up
    return self.update(self.terms[index], vals, msgs, coeffs)
  File "/tmp/venv/local/lib/python2.7/site-packages/trueskill/factorgraph.py", line 145, in update
    pi = 1. / sum(coeffs[x] ** 2 / divs[x].pi for x in xrange(size))
  File "/tmp/venv/local/lib/python2.7/site-packages/trueskill/factorgraph.py", line 145, in <genexpr>
    pi = 1. / sum(coeffs[x] ** 2 / divs[x].pi for x in xrange(size))
ZeroDivisionError: float division by zero

[sublee]

@konikos I've been waiting for any failure cases. I'll review this problem soon. Thanks!

[sublee]

I checked another implementations. But,

1. TrueSkill Calcurator by Microsoft doesn't support more than 8 players.

2. @moserware's C# implementation also cannot calculate 1st case:

   p1 -> μ=NaN, σ=3.7289
   p2 -> μ=NaN, σ=3.8429
   p3~p17 -> μ=NaN, σ=0.5069
   

[konikos]

Ok, nice to know but is there any simple workaround?

[sublee]

@konikos This issue isn't closed yet. I'm still trying to get a nice solution.

[sublee]

This issus is fixed at version 0.1.4. I also added the cases into the test suite. Here's new result of first case:

p1 -> μ=42.234, σ=3.729
p2 -> μ=43.290, σ=3.843
p3~p17 -> μ=16.667, σ=0.507