`rate` in case of a draw (1vs1)

[vwxyzjn]

Hi this is an awesome package and really helpful. Quick question: in the case of draw (1vs1), how should we use the rate function?

In the trueskill, there is a rate_1vs1(r1, r2, drawn=True) feature to rate the skill of the agent. Upon further inspection, the drawn=True manipulates the ranks as follows:

    ranks = [0, 0 if drawn else 1]
    teams = env.rate([(rating1,), (rating2,)], ranks, min_delta=min_delta)
    return teams[0][0], teams[1][0]

Could I handle the draw situation (1vs1) by overriding ranks like done in trueskill? What if I have a draw situation for 2vs2?

Also, see

from openskill import Rating, rate
a1 = b1  = Rating()
print(rate([[a1], [b1]], rank=[0,0]))

from trueskill import Rating, rate_1vs1
a1 = b1  = Rating()
print(rate_1vs1(a1, b1, drawn=True))

[[Rating(mu=27.63523138347365, sigma=8.065506316323548)], [Rating(mu=22.36476861652635, sigma=8.065506316323548)]]
(trueskill.Rating(mu=25.000, sigma=6.458), trueskill.Rating(mu=25.000, sigma=6.458))

Thank you!

[philihp]

Correct.

> r
{ mu: 25, sigma: 8.333333333333334 }
> rate([[r],[r]], { rank: [1,1] })
[
  [ { mu: 25, sigma: 8.065506316323548 } ],
  [ { mu: 25, sigma: 8.065506316323548 } ]
]
> rate([[r,r],[r,r]], { rank: [1,1] })
[
  [
    { mu: 25, sigma: 8.17755635771097 },
    { mu: 25, sigma: 8.17755635771097 }
  ],
  [
    { mu: 25, sigma: 8.17755635771097 },
    { mu: 25, sigma: 8.17755635771097 }
  ]
]

@daegontaven do you think we should add a drawn convenience flag? I don't think it adds too much, but it would increase interoperability with TrueSkill libraries.

[vivekjoshy]

@vwxyzjn It's a bug in openskill.py that passing [0, 0] as ranks doesn't result in draws. I would suggest you pass values starting from 1 until the next release fixes it. But to answer your question, as @philihp already pointed out, simply pass equal value ranks and it will consider it a draw.

@daegontaven do you think we should add a drawn convenience flag? I don't think it adds too much, but it would increase interoperability with TrueSkill libraries.

@philihp I think it's a good idea to have interoperability to allow easier migration. Right now TrueSkill uses separate rate functions for 1v1 and generalized ones. However, it seems only rate_1vs1 has a drawn parameter. If we are going to implement it, I assume it's going to count the number of teams (n) and do an n-way tie as it's ranks.