Glicko ranking bugs

[sann0638]

Great - thanks. Worth a line in the page that says top 25000 are shown, or similar? Saves people like me asking the question

[mubo]

Great - thanks. Worth a line in the page that says top 25000 are shown, or similar? Saves people like me asking the question

Good idea- I'll pop a note up.

[Pipey]

Love the Glickos. Thanks to Nick for all your efforts.
A query rather than a bug (please move / address elsewhere as approp)... but I wonder if there could be any functionality to see your change month-by-month?

Horizontal space is at a premium, but I could add rating change instead of race rank?

This sounds good if you think it fits and will work. Thanks

[kyrre]

Sql-database (sqlite) with Nick's glicko available over here: http://ghoulhq.com/nafdata/export/

I can make all 38364 glicko ranks available in CSV/Excel if anyone is interested.

[mubo]

I've finally tracked down an annoying issue. Sometimes, the change since last month doesn't work properly. This happens sometimes when the games are reported to have been played on the 1st of a month when they haven't been:
eg https://member.thenaf.net/index.php?mod ... advanced=1

For the moment I am just excluding these games until they can be fixed in the db.

[mubo]

I've added new columns that should give a better idea of performance over the last month. Happy to hear feedback on this.

[cyg-Obsidian]

Hello,

I saw you discussed about the fact that Pro elves can't appear without DE and WE.

Have you find a solution for this ?

thanks !

[mubo]

Yes- I can rename them to Elf, or Pro Elves, or pro elf.

Dwarfs/Chaos Dwarfs are the one I don't have a solution for yet.

[mubo]

So to answer your question, yes I can do it. But I haven't implemented it yet. I'll get to it next time!

[cyg-Obsidian]

Thanks for your answers !

I don't know how the DB behind is done but :
what about giving an ID to each race and one table to link this ID to the race name ?

Of course, if you use only one table... it doesn't work.

in that case, do you have the hand on the "search tool" ?

Can you force him to put the search mode to "=" instead of "contains" (as it seems it is now) ?
and add some "jokers" (like * or % )

[mubo]

Thanks for your answers !

I don't know how the DB behind is done but :
what about giving an ID to each race and one table to link this ID to the race name ?

Of course, if you use only one table... it doesn't work.

in that case, do you have the hand on the "search tool" ?

Can you force him to put the search mode to "=" instead of "contains" (as it seems it is now) ?
and add some "jokers" (like * or % )

It's the software that we use to host the table that's the issue.
Initially we had it in "checkbox" search mode, but that caused issues for some users, so we have fallen back to text search only. The tool doesn't support wildcards/jokers.
Turning on exact matching, so "Wood Elves" will give the results, but "Wood Elve" won't. If that's what people would rather have I can do that?

[Anubis]

I have some questions regarding the Glicko System (which I love btw).

Maybe let’s start with “rating” and “decay”. Every coach starts with a rating of 1500, right? How is decay integrated into the ranking? I would like to know how these two very different rankings come into play. Let’s look at Oventa (NAF 9610) & fex (NAF 20993) for High Elves.

Fex has 946.1 ranking with -0.8 decay. He played 2/0/3 in July 2016 for an ELO ranking of 145.01. Oventa has 1474.0 ranking with -1.3 decay. He played 4/1/1 in October 2016 for an ELO ranking of 163.67. Why does Oventa have more decay than Fex, who paused his Elves even longer?

[mubo]

I have some questions regarding the Glicko System (which I love btw).

Maybe let’s start with “rating” and “decay”. Every coach starts with a rating of 1500, right? How is decay integrated into the ranking? I would like to know how these two very different rankings come into play. Let’s look at Oventa (NAF 9610) & fex (NAF 20993) for High Elves.

Fex has 946.1 ranking with -0.8 decay. He played 2/0/3 in July 2016 for an ELO ranking of 145.01. Oventa has 1474.0 ranking with -1.3 decay. He played 4/1/1 in October 2016 for an ELO ranking of 163.67. Why does Oventa have more decay than Fex, who paused his Elves even longer?

So, the last two columns refer to the last month only. So, question is why in the last month has Oventa decayed more than Fex?

It's because Oventa's phi (uncertainty or variance) is lower. The incremental increase in phi (which is what causes decay) plateaus, see the graph in https://www.thenaf.net/rankings/glicko-overview/. Oventa's lower phi could be for a few reasons, a) he has played more recently so has had less time to decay b) the players he played in those 6 games were more established players so we are more confident of the impact of his results or c) he has played 6 games not 5, so more opportunity to lower phi.

Rating is calculated as mu (raw rating) minus 2.5 times phi (variance). Mu term doesn't change over time, while phi increments, causing decay.

Let me know if that helps?

[Anubis]

Hmm this helps a tiny bit, but confuses me as well. The older version of the Glicko ranking showed phi itself, which gave more information to me. Because right now, all I see is decay an d I have no idea who this is calculated. Also I’m confused why Skaven (NAF 19679) decayed more (played 3 games 22.6.2019) than my Elven Union (played 3 games 04.6.2019).

Seems like I just have to check the rating now. I liked it better to see how much phi everyone has, to check the most active players for a race.

[mubo]

Glicko still the same, just the presentation.

I've added new columns that should give a better idea of performance over the last month. Happy to hear feedback on this.

Due to horizontal space, when I added the last month +/- rating and +/- decay I removed the raw numbers of mu and phi. I see your point though, about phi providing extra information. For clarity the decay column is your last month change in phi * -2.5.

Ideally we would have a more interactive solution that could give you all the information. If someone is technically minded and has some time to commit we could do this. But I guess from your perspective you'd rather see mu/phi than last month's change in mu/phi?