Ocean of Code - Alternate ranking

Hi all,

Despite being perfectly happy with my current #1 place at the top of the OOC leaderboard, I wanted to evaluate whether a different ranking system would provide the same results. I present below the results of my analysis.

The analysis is based on the same data that was generated by CG during the rerun. If a player had a lucky streak during the re-run, that player’s score should similarly be inflated in this alternative ranking. No CG servers were harmed in the process :innocent:

Observed winrates

The figure below shows the observed winrates during the re-run between pairs of players.
It is well known that CG picks opponents within a ±5 rank region, hence the data is grouped on a diagonal axis.
image

Expected winrates

I do not intend to go into details on the method in this topic (PM if interested !), but I have applied a technique which reconstructs a score for each player from the observed winrates shown above.

It is possible to derive an expected winrate between two players from these scores.

The figure below shows the expected winrate between all pairs of players within the Legend league.

image

One observation I find interesting from this figure is the four zones that are relatively clear cut.

  • There is a homogeneous 4x4 square at the top left : the top 4 was very well matched.
  • Then comes a second group of 20 people also very well matched down to wlesavo.
  • Then comes a third group with 40 people, relatively homogeneous.
  • And finally comes YoBo, in a league of his own (sorry :sweat_smile:)

Upsets

So… who pulled you down the ranks ? Who pushed you up ?

Here is a list of the biggest upsets in decreasing order of importance.

trictrac         won 88 % against Zenoscave,       he was expected to win 53 %.
jolindien        won 62 % against ValGrowth,       he was expected to win 49 %.
jolindien        won 41 % against siman,           he was expected to win 57 %.
jft63            won 26 % against blasterpoard,    he was expected to win 52 %.
eulerscheZahl    won 33 % against zasmu,           he was expected to win 54 %.
pb4              won 64 % against kovi,            he was expected to win 53 %.
siman            won 39 % against Swagboy,         he was expected to win 57 %.
Zenoscave        won 74 % against tomatoes,        he was expected to win 50 %.
jolindien        won 60 % against kovi,            he was expected to win 52 %.
Neumann          won 36 % against dolmen1234,      he was expected to win 57 %.
ThomasNicoullaud won 76 % against BlueGhost31,     he was expected to win 52 %.
pb4              won 42 % against ValGrowth,       he was expected to win 50 %.
Saelyos          won 37 % against wala,            he was expected to win 51 %.
ameler           won 72 % against Zhmyh,           he was expected to win 53 %.
jolindien        won 45 % against Skril,           he was expected to win 63 %.
daaskare         won 69 % against blasterpoard,    he was expected to win 54 %.
mlomb            won 77 % against trictrac,        he was expected to win 52 %.
Fangel           won 30 % against Illedan,         he was expected to win 49 %.
bourgeof         won 67 % against wlesavo,         he was expected to win 51 %.
gamoul           won 34 % against tomatoes,        he was expected to win 50 %.
Skril            won 40 % against jft63,           he was expected to win 55 %.
kovi             won 68 % against Saelyos,         he was expected to win 59 %.
BorisZ           won 34 % against chucknorris,     he was expected to win 53 %.
trictrac         won 35 % against darkhorse64,     he was expected to win 52 %.
Skril            won 65 % against Swagboy,         he was expected to win 51 %.
Zhmyh            won 68 % against YoBo,            he was expected to win 56 %.
GiB              won 26 % against eulerscheZahl,   he was expected to win 52 %.
Fangel           won 33 % against blasterpoard,    he was expected to win 49 %.
sash             won 33 % against BlueGhost31,     he was expected to win 51 %.
Nagatwin         won 70 % against Nagrarok,        he was expected to win 50 %.

Or, with a visual presentation :
image

Thanks to @eulerscheZahl for providing the data to analyze :slight_smile:

Ciao !

@JBM: it’s coming…
image

11 Likes

Interesting post but i have a question.

In a puzzle like OOC (many hidden informations and a big branching), we already do that we can have a “shifumi” effect, even in the top 20. Is there any danger of assuming an expected winrate in a puzzle like this one ?

Nice analysis, however I don’t think there is enough matches between two specific players to deduce anything.

As far as I know, no ranking system is able to recognize situations such as A > B > C > A. All try to evaluate some sort of “player strength” (ELO) or “player strength distribution” (TrueSkill) associated to the player, not the matchup.

EDIT : Kemeny-Young rank-based estimation does not evaluate a “player strength”, but it is still strongly susceptible to sampling disparity in the case of A > B > C > A situations.