I think the 13k default rank is doing harm [Closed]


#181

The value for a win against completely new player is lower than that for a win against an established player.
The problem is that win and lose against established players with the same rating as you cancel out, while against new players you get then more points for a win as for a lose. This will inevitably result in a small push up for each game against a new player.


#182

Yes, but that is no different from the current system, so I fail to see how humble rank will exacerbate this.


#183

tl;dr the current system matches new players with established players with ≈1500 rating points. With humble rating the match up is agains ≈1148 rating points.

Currently new players get matched up with players around 1500. There you get as many points for a win as for a lose. The rating system assumes new players as 1500. (12.3k vs 12.3k) Since the probability of equal strong players to win is 50%, win and lose are valued equally.
With humble rating, the match up would be 1500 to 1150. We treat new players as 20.6k but the rating system still treats them as 12.3k. Taking the uncertainty of new players into account, it still estimates the chance to win of the established (weaker) player to 21%. Therefor the award for wining would be much bigger than for losing.

The award for winning would be 30 points, for lose the point reduction would be -8 points. (assuming deviation of 100 for the established player).

(ranks are as now and just for illustration)
(with humble rank OGS would show the rank of new players as 20.6k ± 8.3 instead of 12.3k ± 6.6, but the rating points are in both cases 1500 ± 350 )


#184

That’s only against established players… Humble rating is for matchups, not calculations.


#185

Thanks for the great explanation: I hadn’t appreciated the argument for how changing the matching could affect rank increase at all. This explains the concern of Anoek that I didn’t get.

I think the point is that a new player with rating 1500 (and high uncertainty) will now be matched on average with an existing player of 1100, instead of being matched on average with someone at their own rating.

This means you can argue that on average, the person that they are matched with will beat them and gain more points … more.

This effect must happen to a certain extent now anyhow, given beginners will try to match down and dans will try to match up. But it is an interesting point.

I guess that’s what Anoek’s simulations will tell us about: whether it matters, and to what extent.


#186

I suppose that the test is worth looking at, since this is part of what RD intends to counter.

I don’t imagine switching to WHR (which would solve this specific issue) is in the cards (although I’d like to see it happen)

It might even be possible that WHR+chosen ranks would eliminate the main issue of the thread.

At any rate, I’d like to see some sitewide testing for the solution suggested in the unstable ranks thread as well.


#187

Well, WHR is about 345x slower than Glicko so that may be an issue (though Remi would tell us we could still do 1000 games per second).


#188

When you’re all talking about how to pair beginners with other, please keep in mind that OGS has actually 4 different match-making systems in place.

  1. Direct challenges from player to another, these work as they should and not should be changed at all.
  2. Tournaments pair participants together based on how the tourney is set up.
  3. Open challenges, where every single ogs user is free to create a challenge in any way, shape or form he or she wants to. And every user is free to accept any of these challenges if they agree the terms of the match and are within the desired rank pool of the player who posted the challenge.
  4. Quick match finder, for those users who just want to get a game ASAP without caring who will be the specific player they’ll be paired with. You can already tinker a bit on automatch settings, like excluding all players who are more than ±1 stone from your rank, but this option is still made for just getting a game fast.

I guess you’re only talking about pairings from the quick match finder should be done?


#189

I think we’re talking about wherever rank plays a role in pairing.

In all 4 cases you listed, the people you can play with, and the conditions under which you play them, are affected by the rank difference between the players. Humble rank will make it more natural, likely, and even for beginners to play each other.


#190

Of course, there is the added problem that you can’t have ranked games outside nine stones of your rank.


#191

That’s not a problem, really.

If there is a legitimate 9+ stone difference (and I myself would lower that limit to ~5 stones for non-provisional players), making it ranked is nonsensical because the lower ranked player simply won’t win. Anything beyond 3 stones is a teaching game.

If there is no legitimate 9+ stone difference, the lower ranked player is necessarily on a new account or one that hasn’t been used in years, either way they’ll have no trouble getting to their real rank. I tested this before and within 4 games I was at 2d.


#192

I’ve run some numbers and I’m excited to try out the humble rank proposal. It’s going to be a wait and see kinda pattern to see the real effects on the rating system as a whole, but I’m optimistic it’ll at least result in more quality matches overall, and be a lot more beginner friendly during those crucial first few games. We might have to fiddle with the rank bands in the future to account for the difference in point injection into the system, but so long as it improves the quantity and quality of the games played on the server and gets more beginners to stick around and fall in love with the game, then the rest doesn’t really matter and we can figure it out :slight_smile:

This chart indicates to me that currently after about 10 games we have a pretty good idea what someone’s rank is. After about 25 games played, the deviation is pretty flat all the way out as far as you want to measure.

What I’m thinking is we implement @Tofoon’s humble rank proposal with the modification that we interpolate the humbleness adjustment based on deviation between 350 and 125, so that after an expected ~12.5 games played, we’re using your true rating. That is to say,

humble_rating = rating - ((min(350, max(125, deviation)) - 125) / (350 - 125)) * deviation

In practice what this means is that someone who just joined will have their 1500 ± 350 rating as they do now. For the purposes of matchmaking though we’ll consider their rating to be 1150, which corresponds to 21k. Assuming they play another 21k’er, if they win they go to an effective 17k, if they lose they go to 25k. Of course if you pit two brand new players against each other, the winner is going to go on to challenge at an effective rank of 14k, which might be a bit high if it’s their second game ever, but at least they’ll have hopefully had a positive experience with their victory in their first game so when they get crushed in their second it doesn’t hurt so bad and they know it is at least possible to win this infuriating game :slight_smile:

As you continue to play games the effects of the humble rating adjustment dissipates as your deviation goes down, and after about a dozen games or so the effects will have gone away completely and we’ll be using your true rating to do matchmaking.


The following chart doesn’t have any real relation to anything in this post, but it’s fun to look, and it serves as a benchmark for us heading into this proposed change - hopefully the red line will go up afterwards :slight_smile: . One thing to note though is that despite the problems noted in this thread, it’s still the best system we’ve had (at least when considering the number of games new players are playing, which I think is an important metric). And if the humble rank produces better matches for beginners and ~13k’ers, it should only get better.


#193

What about let new players choose what rank to display next to their names?
Just display, nothing rank calculation related. [25k?] or [9d?] instead of just [?] as it is now.
Most people will not lie. It will be useful in custom challenges.


#194

I think anoek is mostly trying to make it easy for new people to come here, and new people don’t know their rank.


#195

Yep, I think it’d be a point of confusion, and it’s usefulness would vanish after a few games anyways. Established players wanting to do custom challenges will likely already have a rapport with those they’re challenging so having ? vs 3d? next to their name for a few games shouldn’t make that much of a practical difference I wouldn’t think


#196

Thought I’d throw in my experience as a completely new player who recently joined the site. Seeing a lot of speculation about the experience is like for people like me, but not a lot of actual new players reporting on their actual experiences.

For context, I have long been generally aware of the basics of Go, but never played a game or looked that seriously at it until about 2 weeks ago. Found this site almost immediately, spent some time looking at introductory resources, poking at life-or-death puzzles, and watching games; then I settled into playing some quick 9x9s against Kugutsu, a lot of forfeiting early, just trying to get the start of a sense for how things actually play out. Finally took the plunge and started doing ranked games a couple of days ago, still 9x9. I opted to create custom games rather than just quickjoin someone else’s - still 9x9, a few blitz but mostly live, ranked, japanese scoring, otherwise default settings.

Other than one game that took about 2 minutes, had someone accept every match in under a minute. After three losses I was at 24k, played 10 total matches now and after dropping to 25k, my first proper (not due to time-out) win has me sitting back at 24k currently.

Anyway, single sample of anecdotal evidence, but for myself, I don’t feel like I was hurt at all by the current system. Sure, I’ve had some punishing defeats, but some of my more recent losses to other ~25k players have been just as crushing as my early losses to higher-kyu players, and out of a mere 10 games, I’ve managed one proper win, and had another very close game that, on review, I was able to find a way I could’ve turned it around at the end and won.

If a lot of people are quitting after their first game, I personally suspect this is a fact about them and about Go in general, and not suggestive of any flaw or failing in OGS. I just can’t imagine how you could ensure everyone’s very first game is against an equally-skilled (or unskilled) opponent. If you could somehow ensure most of them come out of their first game feeling confident, like they could win next time if they didn’t this time, it might increase the amount of people who play at least 5 games, but I doubt it would significantly increase the amount who actually stick with the game long-term and become invested in the game and community.

anyway, that’s the perspective of one actual novice learning on the site.


#197

Thanks for taking the time to write about your experience! Glad to hear the current system wasn’t too off putting :slight_smile:


#198

My experience was very similar, except I used automatch for my first game and got matched with a 12k player. I cancelled before the first move and then visited chat for advice. They suggested playing anyone and just resigning fairly quickly to get my rank down. I actually decided to use a bot for this purpose.


#199

That’s an interesting point. Because when I first got in touch with go in 2014, under old system, I played 6 ranked and 1 unranked game and won 4 of them. And I still quit this game :stuck_out_tongue: Took a long time to rediscover it.

Here’s a genius solution from me. It’s only a problem if new player annoys stronger player on 19x19 board. How about restricting new players to 9x9 board until they prove their rank? I know it’s bad, just wanted to write it.


#200

I was planning on making the default automatch settings 9x9 when your a new player, but I wouldn’t want to restrict that per say, just encourage it by default.