Messages

I analyzed the full leaderboard at Scavenger (http://dominion.lauxnet.com/leaderboard/) and I noticed that the initial phi=2 seems to be chosen too high. Compared to chess, luck just makes it more difficult to beat opponents consistently – and there are fewer pros. So currently 95% of players (with at least 20 games) have a mu between [-1.95,1.68]. Whereas for a new player it is implicitly assumed that the 95% range is [-4,4].
This gives the odd results for some players with few games who get to very high/low mu. Currently, if someone new just beats a mu>2 player once, they'll end up with a mu around 2.5, which just doesn't sound right, given the luck involved in Dominion. (the change in mu is approximately phi^2*(wins-expected wins))

Therefore, I'd suggest to lower the parameter to phi=1 for new players and also to cap phi there. (Capping was suggested in glicko1, and it seems reasonable to me that someone who has played shouldn't have a higher rating deviation than someone new).

After a couple of months, it would actually be possible to estimate the parameters for initial phi, sigma, and tau, that give the best results in predicting game outcomes.


I also have some thoughts on the definition of a good match. I think comparing levels is bad especially at the lower end of the leaderboard. There are some people with a very low mu and high phi, which results in low levels and makes them a supposedly bad match for many opponents. Whereas actually we are very uncertain that their mu is that low.

My preferred way would be to use expected win probabilities and let players set a range. The advantage would be that a certain winning probability is more understandable for the layman than some level difference.

If you want to keep a criterion closer to the current system, I would define the range of suitable opponents as [mu-phi-x, mu+phi+x] with some cutoff x (x=0.5 seems reasonable to me). That would mean that there are more possible opponents for a player with a high phi (the system doesn't know the skill well) than for a player with a low phi (good estimate of the skill).

A more sophisticated matching algorithm could also check the distribution of players that started a match say in the last 30 minutes to determine a good cutoff. (When there are more players and/or a player is more in the middle of the distribution, you can find a more equal opponent within a certain time than for players in the tail of the distribution.)

Everybody is missing those days.
I don't understand your other question. Obviously, that is a bug that shouldn't happen. But you can just pretend that these two days didn't exist and the system can continue to "work" on from there.

Still there seems to be a mismatch between the games I played according to the in-game client and that website linked to by tufftaeh.

According to the game client, I have played 24 games now (2017/04/23). That number was 23 games when I had last checked in, 2 days ago. I played one game on that day, so that would make sense. But why am I then finding that I have currently 28 ranked games when looking at lauxnet.com?

I think the official leaderboard misses the games from last Wednesday and Thursday. In your case, those are the 4 games currently marked as 5 days ago in "Scavenger". Therefore, the official count is 4 lower than the sum of games on Scavenger.

The tab hasn't been updated, but you see the new level, when you start a game. You can also check your progress here:
http://dominion.lauxnet.com/scavenger/?user=katie_mi

I would prefer to have the definition of an equal opponent being narrowed down. The set of possible opponents for players with a high level is obviously skewed towards getting weaker opponents, but I would often rather wait (much) longer than getting matched again with some >80% expected win opponent. (My average matched probability has been around 75% last week.)

I think it would be fine to reduce the threshold for a suitable opponent to 7-10 levels. Ideally, I would like to have a slider that sets the range of accepted win probabilities. But (another) button with "more equal" opponents could work as well, so that people have the choice between random, decent, and equal opponent.

When gaining a Villa with Watchtower in hand, one has to select "Villa" from the log to put it in hand. I think it should say "may put it in hand" instead. (Ideally this should have a button as well.)

Enchantress of course - I should have loaded the game myself.  ::) In earlier possessed turns, I had remembered that Enchantress was in place. In my defense, my opponent didn't get what was going on either.  ;)

What made it difficult to catch though, is that the log is "destroyed" (scrollbar/earlier turns missing) after possessed turns, such that I couldn't see "buys a Villa" without reconnecting.

In game 2883855 the last turn I possessed ended abruptly when I gained a Villa with University.

The only good news for you is that those failed attempts don't count for your rating.

1) Visibility of opponent's duration attacks and -1 coin/card tokens.

2) Being able to see what exactly the duration cards /set aside cards are. For example, with Throne Room it's not always clear to me which duration is throned. For Haven and Archive I would like to be able to see which cards are set aside / how many cards my opponent has set aside in each Archive. (I'm not sure whether the rules technically allow you to look at the cards in each Archive other than at the start of the turn, but with a complete log available, it's just a pain to go back to the last turns and figure out what cards are in the Archives.)

3) Reduce necessity of having to look at / use log.

I'm reviving this topic, because I'm back at the same location as in January and experiencing the same connection problems again with Firefox, whereas I don't seem to have any troubles with Chrome. The problems seem to be the same that DG is experiencing - he mentioned also using Firefox.

With the same laptop (Windows 10) I didn't have these problems at other places. The internet connection is fine for any other purposes.

I'm attaching the screenshot from the console. Let me know, if there's any other diagnostics that I can do to help you solve the problem - I'm here for the next 10 days.

I agree with Stef. Ratings are primarily used to improve matching when seeking games. There is a danger that including the AI as a player would mess this up more than it helps right now (As Stef mentions, we have seen this in the past).

I think it's the opposite: I can see that you don't want people to claim top spots in the leaderboard, when they are only/mostly playing bots. But for matching quality it should help to also use the data from bot games, and the possible downside is quite limited: if the rating coming from bots is really wrong AND this player decides to play a human at some point, he'll lose the game(s) and quickly lose in terms of rating. I think the leaderboard / match quality gets more distorted by not all games being played with all expansions.

Once you have bots of different quality, however, this rating would also help to get the best suited bot for those players that only play bot games.

So I think it would be fine to have an additional rating that takes into account both bots and human games.

I think that you get rid of some of the potential problems mentioned by only using games against bots, where I explicitly say that I want a rated game and I don't have any influence on the kingdom other than through  my subscription.

Then, people might not care less about finishing that game properly than against a human opponent. And there are actually uspides (e.g. if you have to stop playing immediately, you can finish later, whereas it would be a loss against a human opponent; and the bot never has connection troubles / slow plays you).

If you want to be conservative, you can reduce the weight of bot games, e.g. 10 wins/losses against the bot count the same as 1 win/loss against a similar ranked human.

Rating inflation due to newcomers is a valid concern, although it has some upper bound as well. (With the current settings you can't get above mu=3 even with a 100% win rate against a lot of mu=0,phi=2 players. And that is an extreme case, because the bot is assumed not to lose (to better players). I expect the top 20 to move towards mu=6 instead.)

I think the most valid concern is the getting to know and exploiting the bots general weaknesses. At the moment this wouldn't matter at all for the top of the leaderboard. But if the bots get better and rise in their rating this skill could be valuable but shouldn't improve your position on the leaderboard (e.g. if I knew that they don't go for (more or less complicated) 3-piles, I could adjust my play and win more often than against some random opponent)

I'm adding this thought to the "general discussion" of leaderboard: I would allow to have rated games against the bot, if they are drawn randomly (like in "find a game"). The bot would then also get a rating - and could be shown on the leaderboard.

Personally, I only end up in bot games now, if I load a game or want to try something out and typically resign on them. And of course they shouldn't be rated. Also people shouldn't be able to select specific kingdoms to game the bot.

When the bot is not great like now, it will just have a low rating and beating it will not get you to the top of the leaderboard. But apparently there are some people who only/mostly want to play bot games and you could provide them with a rating, which is more informative than level 20.

And you would get feedback on your future bot improvements, when its rating rises.  ;)

It's true that you want to have the rating period at least one day (in the paper it says best at 10-15 games per period). But it would be possible to show an update during the day, which would be the (exact) prediction of your level at the end of the day, if you stop playing now. For details see my post here.

I kind of like that there's one published leaderboard per day. And my suggestion above would not affect the ratings at all (e.g. whether you first win against someone and then lose or the other way around is irrelevant within the rating period). But I can see that some people want to get some instant feedback.