Quest for a rank formula

Lots of people have asked us on the rank formula, so here it is. We would really appreciate it if you helped us with ideas or told us your opinion on how you think the rank should be computed.

Criteria for a good formula :

1. more wins -> increase the rank

2. more defeats -> lower the rank

3. new photos -> neutral rank

4. the tops should not bounce; newcomers should not reach the top too easy.

In the picture below you can see the rank, resulted from the number of defeats and wins. On the horizontal axis, we have defeats / battlesMax and wins / battlesMax (where battlesMax is the maximum number of battles ever reached by a picture)

The way it works :

- a new photo enters the rank of 50%.

- with every defeat, it goes along the blue line (lower rank)

- with every win, it goes along the orange line (higher rank)

- it takes a while for the newcomers to reach the top (this is natural because it takes time to raise enough votes to enter the top)

rank1.png

4 Responses to “Quest for a rank formula”

  1. Diane, flickrcombat.com Says:

    BattlesMax is the maximum number of battles occured in the certain category since the system has been online.

  2. joe Says:

    Cool site, nice execution. A couple of suggestions.

    Having the rank as a percentage is not good. It makes it seem like it’s a percentage wins figure, or a percentile, when it doesn’t seem to be either, right? If it’s just an arbitrary number, don’t use a percentage. Also, the spread is very small — the most popular shots are in the 50s, the worst ones are in the 40s, that’s not much variation.

    Also, a “worst 21″ listing would be great.

    Finally, the “x% of people agree with you” figure is frustrating. Most of the time it is 0.0%, presumably because the pair hasn’t been compared by anyone. But it might be that 0 of 3 people disagreed. The two cases aren’t distinguished. Better would be something like “3 out of 4 people agree with you”.

    Thanks!

  3. Diane, flickrcombat.com Says:

    This idea you mentioned last is actually very good, and we’ll give some thought to it. Thank you so much for your feed back.

  4. joe Says:

    You didn’t seem to like my comment about the ranking.

    Currently the #2 babe has 483 wins, 222 losses, for a 68.5% win ratio. Her rank is 60.8%.

    The #4 babe has 211 wins, 25 losses, for a 89.4% win ratio. Her rank is 57.7%.

    The #4 babe is clearly doing better, but she’s being punished for not having been involved in as many comparisons. This is not fair. Think about a picture with 100 wins, 0 losses. It should have a perfect score, but won’t; it will even have a worse score than a picture with 300 wins and 100 losses.

    I’m not trying to be annoying, I’m just trying to help give fairer results :)

    I suggest that you just go for a win/loss percentage ratio. This is fair to all pictures, regardless of how many comparisons they’ve been involved in. However, it’s prone to unreliable results for pictures with very few comparisons, so don’t give a ranking until a picture has been involved in a certain number of comparisons, eg. 10 or 20 or 50.

Leave a Reply