question

Benjamin Golinvaux avatar image
Benjamin Golinvaux asked

Leaderboard position estimation : request for info

Hello

We already know that leaderboards (v1) are unable to retrieve the exact ordering of players in a leaderboard and only lead an estimate that is consistent when called from a fixed player (so that a given player always see the same ordering in the leaderboard they retrieve)

However, I have some additional questions :

• I seemed to recall that the 1000 first entries are guaranteed to be perfectly sorted. However, we have far less than 1000 players (we're in the testing stage and this particular leaderboard has slightly more than 100 entries) and we see differences based on the playfabID performing the GetLeaderBoardAround... call. Is this to be expected?

• I also recall someone from PlayFab explaining that the estimation algorithm works best with normal (Gaussian) distributions. What happens if our distribution is not symetric around the mean? Should we maybe use a fixed transformation of the data before storing it so that we try to make it normal?


Thanks for the light you might shed on these issues.

Best

Leaderboards and Statistics
10 |1200

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

1 Answer

·
Xiao Zha avatar image
Xiao Zha answered

>>I seemed to recall that the 1000 first entries are guaranteed to be perfectly sorted. However, we have far less than 1000 players (we're in the testing stage and this particular leaderboard has slightly more than 100 entries) and we see differences based on the playfabID performing the GetLeaderBoardAround... call. Is this to be expected?

If you have the same statistic value in a leaderboard, when you call GetLeaderboardAroundUser API with PlayFabIds that are on both sides of the players with the same statistic value, the positions of the players with same statistic value are different and this is by design. If this is not your case, could you provide the Leaderboard screenshot, the different results of the API and your TitleId for us to research.

>> I also recall someone from PlayFab explaining that the estimation algorithm works best with normal (Gaussian) distributions. What happens if our distribution is not symetric around the mean? Should we maybe use a fixed transformation of the data before storing it so that we try to make it normal?

You may refer to Leaderboard positions wrong / duplicated when viewing multiple leaderboards - Playfab Community, as the Brendan said, if your distribution is not symmetric around the mean, for example, like a diminishing result curve, you will likely to see more variance in the values that are past the top 1000, and this won’t affect the statistic value within 1000. In addition, Since players are highly sensitive to position when they’re near the top of a leaderboard, they’re not when they’re no where near the top. So, there is no need to normalize the data.

2 comments
10 |1200

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Benjamin Golinvaux avatar image Benjamin Golinvaux commented ·

Thanks for your explanation.

You thus confirm that, in the case where all players have different statistic values, the first 1000 are guaranteed to be sorted correctly, whoever makes the GetLeaderboardAroundUser call, do we agree?

0 Likes 0 ·
Xiao Zha avatar image Xiao Zha Benjamin Golinvaux commented ·

Yes, when all players have different statistic values, the first 1000 are guaranteed to be sorted correctly.

1 Like 1 ·

Write an Answer

Hint: Notify or tag a user in this post by typing @username.

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.