To reproduce: https://gist.github.com/975955
Interestingly, I run the same program using the version by @dmj111 which removes the need for scipy https://github.com/dmj111/trueskill , and there it throws the following (line numbers might be slightly inaccurate as I edited the files a bit):
...
File "/something/trueskill.py", line 191, in _InternalUpdate
new_pi = 1.0 / (sum(a[j]*a[j] / (y[j].pi - fy[j].pi) for j in range(len(a))))
File "/something/trueskill.py", line 191, in <genexpr>
new_pi = 1.0 / (sum(a[j]*a[j] / (y[j].pi - fy[j].pi) for j in range(len(a))))
ZeroDivisionError: float division by zero
I would appreciate that you at least give me some insight about what is wrong here, since trueskill is quite a complicated algorithm that I don't understand fully...
(by the way, the dataset that I commented out runs just fine with your version but causes ZeroDivisionError in @dmj111 version.)
To reproduce: https://gist.github.com/975955
Interestingly, I run the same program using the version by @dmj111 which removes the need for scipy https://github.com/dmj111/trueskill , and there it throws the following (line numbers might be slightly inaccurate as I edited the files a bit):
I would appreciate that you at least give me some insight about what is wrong here, since trueskill is quite a complicated algorithm that I don't understand fully...
(by the way, the dataset that I commented out runs just fine with your version but causes ZeroDivisionError in @dmj111 version.)