Lichess4545 Ledger 099

Lichess4545 Ledger

Issue #099 - May 16, 2019

State of the 4545 League

by @kostasvl

And.... lift off! The 17th season of the 4545 team league has launched successfully! 28 teams are fighting for this season's trophy. If you are new to the league don't forget to read all of the useful documents that will help you find your steps in this league. For any further questions you can always summon the moderators in #general with the following command: "@chesster summon mods".

Side-leagues News

by @kostasvl

After 8 rounds in LoneWolf things are more clear in the top places of the Open section. @moritex is leading with 7 points and he is followed by NM @ChristopherChabris who has 6.5 points. 4 players are in 3rd place chasing them down. Note that there are still 3 rounds left so anything is possible. Board 1 game of the week is @moritex vs @IsaVulpes.

In the U1800 section of LoneWolf players are even closer to each other. Sole leader in the section is @SinquefieldCup with 7 points. In 2nd place there are 3 players with 6.5 points and they are followed by 3 players who have 6 points. Board 1 game of the week is @SinquefieldCup vs @PGT1.

The 6th season of the 960 league is completed! Congratulations to the top finishers and thanks to the moderators for this season!

1st place: BOT @Vetinari_Computer (The engine of the league, scoring 6.5/7 points).
2nd place: @Somethingpretentious (Who won the previous season scored 6/7 points).
3rd place: @Grzybozbur (Scored 5/7 points).
U1600 winner: @seius (Scored 4/7 points).

You can see detailed standings of this season here.

Last Sunday Blitz Battle 60 was held and 7 participants joined that battle. Congratulations to @mronemore for winning the battle. You can see detailed standings of the battle here. There will also be another battle this week. For more information join the #blitz-battle channel.

The 7th Season of Series started this week and the format has changed. Now players can rise to the top groups faster compared with previous season. For more information about Series join #series-general.

The pool phase of the 5th Rapid battle ended and the 1st round of the elimination stage has started and some games are already played. Congratulations to everyone who qualified and good luck for rest of the battle!

The 1st correspondence tournament that started a couple of months ago, ended with @pushedpawn being the winner. The 2nd correspondence tournament is ongoing and the 3rd one will start soon. To stay tuned about the registration of the tournament join the #correspondence channel.

The May Simul will be hosted by NM @ChristopherChabris on Saturday May 18th at 13:00 UTC. There are no restrictions for this simul. Make sure to join #4545-simul so that you don't forget to join the simul. The simul will open 30 minutes before the starting time.

Interview

by @adande1 and @kostasvl with @moistvonlipwig

Our featured player this week is perhaps most well-known (among other things) for his penchant of getting into deeply technical conversations on Slack, trying to mathematically formulate, prove out and solve every problem in the various leagues, community and the universe in general. Moistvonlipwig (“MVL”) won the Masters Section of the most recent (Season#6) 9030 Chess Series, a league that he co-founded and to this day is also one of the mods. This is MVL’s second win and third top three finishes in the 9030 Chess Series, having previously won Season#1 and placing 2^nd in Season#2. MVL was also on the winning team (We're OK with any name, actually) of the Lichess4545 League in Season #10 and hosted the 4545 simul in December 2018. However, MVL’s most interesting feats maybe the ones accomplished by his very own bot Vetinari_Computer who will be declared winner of the Lichess960 League Season#6 and is also doing quite well in the Correspondence League.

Tell us a little something about yourself. How long have you been playing chess and what has been your history/ participation with the different leagues in the community?

I’ve been playing since I was 6 or 7 years old was quite decent at young age. Following a not so great chess streak I started studying at university and since then am not very active anymore only playing the odd tournament and mostly this league.

Here I played some seasons as main player, some as alt and also some rounds of ladder and later the Series.(as well as some side events like blitz or rapid battles) Overall enjoying my time and the opportunity to keep playing despite rarely having the time to play a full OTB tournament.

Please answer some fun facts below

Favorite opening as White: Young me would probably name the Two Knights defense (1. e4 e5 2. Nf3 Nc6 3.Bc4 Nf6 and 4.Ng5 there) with either colour. Nowadays I am less clear about that but don’t have a better answer to offer.

Favorite opening as Black against 1.e4: See above, e5.

Favorite opening as Black against 1.d4: Probably the Nimzo though I never felt great about my Black repertoire against d4.

Favorite chess icon/ inspiration: Can I name myself five years from now?

Most memorable achievement (preferably chess related): Getting a 50% life time score against GMs (with that one draw this Grenke) is a nice achievement I guess, even though the game was not really memorable at all.

Most memorable failure (preferably chess related): I don’t recall a whole lot so something recent: In the deciding match for the OTB league championship I got my only loss of the season after getting exactly the prepared line and then still messing the moves up.

Most memorable game and/ or opponent (preferably from Series#6) : Playing against mn8 always is a great fight with our first game being (I think) my first game in this community, a ladder match, as well as various Series and other games. (this Series’ game was not super interesting but a nice tactic at least)

Tell us about your username – why did you pick it? What about your bot’s name Vetinari_Computer?

There’s not much behind it, both are characters from Terry Pratchetts Discworld. I can recommend the great book “Going Postal” as further reference.

From some of your conversations on Slack, you clearly enjoy deep technical discussions and solving problems using mathematical formulations. Do you find yourself getting into such discussions in real life as well or do you think this is a function of the tech savvy community we have?

As a mathematics and computer science student those do at times happen away from the keyboard too though less frequently mostly due to me not being away from the keyboard very often.

What role does chess play in your life?

It certainly was a large part of my youth, nowadays it sadly had to step back a bit.

Do you have any goals with respect to chess?

Long term becoming FM would be nice, and IM would be amazing of course for free entry to tournaments alone. However either one will require a lot of work that I am not putting in at the moment. I am considering though to maybe take half a year off at some point and focus on things like improving at chess and working on the bot etc.

Talk about your recent performance and the biggest factors behind your success in the 9030 Chess Series.

Being one of the rating favorite certainly helps. As does the rating favorite mn8 falling for a rather cheap tactic in our game. After that game it was mostly trying to get into the games well rested and if possible somewhat prepared. Finally a bit of luck also doesn’t hurt.

Share your thoughts on the 9030 Chess Series – what changes, if any, would you like to see in the future?

Being part of the mod team I can make the changes that I want to see. For next season we will be trying a different promotion/relegation structure to allow quicker climbing as well as slightly stricter rules to make forfeits happen less often. Let’s see how it turns out in practice.

You recently participated in the Grenke Open - tell us about your experience. Were you happy with the results? How often and how long have you been playing in this or other similar tournaments?

The Grenke always is a tough tournament especially due to the first round being late messing up my sleep schedule. My result and play this year was alright though, scoring a solid 50% in a pretty strong field and not making too many blunders.

I’ve been playing tournaments for most of my chess life, including for at least 10 years now big Open tournaments. When I was still going to school I was playing a lot more of these, nowadays it is pretty much only the Grenke.

Tell us about the history/ origin of your bot Vetinari_Computer? How much time/ effort did it take you to build and what are some of your short/ long term plans for its future?

At this point it’s hard to tell how much time went into it since I rarely work on while the project has existed for a couple of years already. I’ve been wanting to implement some algorithmic improvements as well as teaching it other board games (e.g. Thud once again from the Discworld) at some point. Short term I am not planning on working much on it though.

You probably get asked this a lot but what advice and/or resources can you point to for someone who is interested in exploring building their own bot?

This depends on what they are trying to do. For writing an as strong as possible bot, trying to implement the concepts explained here should be a good road.

I myself was less concerned about the bots strength and more about trying to find out things myself and having fun on the way. In that case of course just start writing and see where it goes.

You had some interesting ideas on the optimal strategy to win the Fantasy League – can you share them with us again? How come you’ve never participated in the League to put the strategy to test? Do you think that SP’s back to back wins is sheer luck or has he figured out the winning strategy?

I haven’t actually worked that much on the topic though I still think a good strategy to reach first place is to build a rather risky line up and hope to get the better side of the dice roll. To participate myself I would like to create a lineup based on some of that theory and backed by some statistics but sadly I couldn’t find the time to do that. (and I didn’t really feel like participating with a semi well thought out lineup)

Due to not following so closely I don’t know how the past seasons went but a mod winning back to back seasons sounds like insider information being used...

Talk to us about your 4545 simul experience back in December where you scored an impressive 8 wins, 0 draws and 1 loss. Have you hosted simuls before and how did you prep for it? Are you looking forward to hosting the next simul?

I don’t recall any simuls before that and I also didn’t really prepare for the simul specifically.

I made a lot of mistakes at the start which probably was due to that lack of simul experience however fortunately in many of the games I was able to still turn it around. So not that happy with my play but the end result is decent.

Of course I would also be interested in some day hosting another simul and try to get better at it.

Describe your chess regimen. What chess resources (books, applications, websites etc.) do you use regularly and how did these help you?

In the past I have done some tactics and a lot of online blitz as well as listening to live commentary of events and of course, actually playing tournaments. These days it is mostly playing here and occasionally doing some tactics on chesstempo. (and I feel like even the limited amount of time I put into the latter helped a lot in the Grenke)

How has this League and the community affected your experience of the game?

Playing here is somewhat similar to playing OTB except much more relaxed (usually not more than one game per day). So I think the experience is similar to an OTB tournament with one round per day which I always enjoyed. I probably also learned a lot just through discussions in various channels.

Any final thoughts you’d like to share with the community?

As mentioned above I am enjoying my time here, hopefully the future seasons will run just as smoothly. :)

Thanks for taking the time and we hope to see both you and Vetinari_Computer continue to do well in the future!

If you liked the featured player interview and have any thoughts on who and/ or what you’d like hear next feel free to post in #lichessledger or DM @adande1.

Project

by @erinyu

How much is the first-move advantage? And are we improving?

Hey, my name is Erin. Didn’t expect to see me here, did you? Two questions.

1) Should you resign if you’re playing black? Well, yes.

2) Are we getting better? Well, maybe.

***

In 2015, Jeff Sonas, with ChessBase, showed that white has an advantage equivalent to 35 rating points in FIDE-rated standard play [1]. However, he notes that the effect probably attenuates with faster time controls and with decreasing player strength. Consequently, this result doesn’t have to match up with our league.

Lichess uses the Glicko2 rating system. Given the difference between white’s rating and black’s rating, , we may calculate the expected value of the white pieces score using the Glicko equation below [2].

The function ' requires knowledge of the two players’ rating deviations, which we need to calculate for every single game. We notice that if and , we get the Elo equation. If , then it’ll get “absorbed” into in any case.

If white is able to use the first-move advantage, then the expected score for both players will be 50% when white has a lower rating than black. As a result, we translate , where is the rating boost due to having the white pieces. If we define the average of to be some other number , we can linearize the above equation and get the one below.

We’ll find experimental values using data from several Lone Wolf and Team League seasons, chop them up into a workable number of averaged data points, perform a least squares linear regression [3], and figure out what is. Additionally, Dr. Glickman finds that, at least for USCF-rated games, [4]. Assuming an average Lichess Classical rating deviation of 70, we will see whether in our leagues.

Dr. Glickman explains the increased is due to low-rated players improving quickly. Another way to phrase this is that when is unusually high, ratings are more variable estimations of player strength. Since lower-rated players win more than the theoretical model expects, it must be the case that either lower-rated players are stronger than their rating, or higher-rated players are weaker than their rating. More likely, weaker-players are just getting better.

Why can’t it be anything else? Maybe players new to the league are over-rated compared to the rest of us (whether it be inexperience at OTB time controls, or just simply closed pool inflation). Well, nobody has posted to the Ledger any convincing evidence supporting this claim. To the contrary, in the 2018 Lichess4545 Anniversary tournament, league veterans who played at least 5 games did not show any significant rating swings on average and in fact lost a little, so it seems unlikely that lower-rated players are beating up on inflated newcomers.

Could players new to Lichess, all of unknown and highly variable skill yet all starting off at 1500, introduce rating variance at lower-levels? Yes, possibly, but reasons like this are why this method can merely suggest our improvement, not show it.

Anyway, though Glickman’s hypothesis has a few confounding variables attached to it, it will still help us determine whether @parrotz and friends are seeing their efforts come to fruition [5]. Onwards.

***

The 45+45 League is a bit troublesome to work with because the mods do an excellent job of preventing rating mismatches, which makes it difficult to collect a healthy range of data while taking averages. Through a few suspect data-sorting procedures, I manage to create a pretty-looking plot. I should say that regardless of how I sort the data, the regression coefficients remain within the standard error.

To be clear, I take all the results, sort them by white’s rating difference (typically a range of -600 to +600) and create data points by averaging every predetermined number of games. The plot below shows the actual, sorted, grouped results as blue dots, the fitted function as the orange curve, and the rating boost as the horizontal distance between the two dotted black lines.

Figure 1. 45+45 League: White rating difference vs. white score.

I calculate a first-move advantage of 37 points, which matches Sonas’ 35-point boost. If we calculate the first-move advantage from purely white’s score, which is 53.2%, we get 31 points, and this falls barely within one standard error. For the 45+45 League, , which is well above the theoretical value. This indicates that ratings are more variable than what Glicko2 predicts, and a very likely culprit is that our players are getting good at chess. Hooray!

The regressions, separated by board number, are not as “well-behaved” as this one, primarily due to narrow sample range and sample size issues; Dr. Glickman works with 225,000+ games for his work, while I only have slightly fewer than 4,800. I won’t show the plots here to avoid spam, but feel free to DM me @erinyu if you’d like to see; the only good-looking, decent-ranged regression derives from board 1. Much to my chagrin, for board 1. Maybe our board 1 members are not improving as much, but it is important to note that we play games every week, while an OTB player may actually have time to improve in-between tournaments every few months. Regardless, Dr. Glickman also finds that rating deviation decreases the higher the rating is, which correctly implies that lower-rated players are not as consistent playing the game of chess itself. In this sense we’re consistent with a legend’s previous work.

The Lone Wolf League, by comparison, is much easier to work with because rating mismatches occur virtually every other game. The mods think of a brilliant idea by creating U1800 and Open sections, yet I don’t have to employ any clever methods to generate the pretty plot shown below.

Figure 2. LW League: White rating difference vs. white score.

The first-move advantage is equivalent to merely 9 points, and this makes sense as the time control is faster. However, I’m not exactly sure how much lower it should be. White’s overall score is 50.9% which corresponds to a 9-point advantage as well. For the LW league, , so ratings are just as variable as they are in the 45+45 League and still suggesting that our players are improving! Yippee! Just as before, if you believe me, the ratings show less deviation in the Open section than in the U1800 section, which we expect.

Separating the search into only the U1800 results, the data has a bit more scatter, but everyone expects more variation among weaker players (not to say board 1 players are any better… I know otherwise first-hand). I calculate a 20-point advantage for white, though the standard error is on the higher side. This surprises me because I would expect the lower-rated players to not use their first-move advantage as well as the higher-rated players in the Open section. However…

Trying this with Lone Wolf Open section results, the rating boost is virtually zero, and I’m not convinced it’s a data-handling fault. White indeed scores 50.2% across over 1,600 games played from Open seasons 10-13, corresponding to a puny advantage, if any at all. Perhaps the prestige of a LW Open first place victory compels both sides to play ambitiously, and the players’ inability to handle the resulting complications dampens the first-move advantage? Or perhaps 1,600 games just isn’t a large enough sample size? Regardless, the model fits the data closely, even if it throws my instincts to the moon and back [6]. I show the plot below because it’s so bizarre to me… Seriously? Zero advantage?

Figure 3. LW Open: White rating difference vs white score.

***

We show that 45+45 players win more often as white, so much so that they perform 30+ rating points better than expected if there were no advantage. The effect isn’t as pronounced in the Lone Wolf League, but still exists somewhat at 9 points. For the LW Open section, there isn’t enough evidence, with this method, to show white has any advantage at all. If anyone has any clues for why this is, please let us know in #lichessledger or send me a DM @erinyu.

Comparing our regressed values to the theoretical , which assumes both players have an RD of 70, they are much larger. So, there is a good amount of variation unexplained by Glicko2 uncertainty alone. This suggests, in a very provocative sense, that our players are improving!

In order to truly determine whether we are improving, we may analyze our league in at least two different ways: 1) test for improving move quality as determined by a strong chess engine, and 2) test for player rating increases over time. The drawbacks for these are high standard errors and possible inflation in the Classical pool, but the results in this article give us confidence that these approaches may verify what we want to hear.

As always, feedback is very much appreciated. I am not a statistician or mathematician. In fact, I’m hardly even an Erin half of the time. I just like investigating these things. Think I should know something? Let me know. I’ll probably inundate you with hearts.

Special thanks to @moistvonlipwig, @toni4127, and @colwem for responding to my DMs.

To Learn More

[1] https://en.chessbase.com/post/the-sonas-rating-formula-better-than-elo

[2] http://www.glicko.net/glicko/glicko.pdf

[3] https://www.youtube.com/watch?v=VqD-nf1YUks

[4] http://www.glicko.net/research/chance.pdf

[5] https://www.lichess4545.com/team4545/season/17/document/league-history/

[6] https://www.youtube.com/watch?v=JIFFQgvxXow

Video

by @okei and @pepellou

From this study @okei his reviewing the games of @Zher0 in the 45+45 Season 16

Stream of @pepellou (testing himself on the hard puzzles of the Ledger #98 at some point)

Weekly Stats

by @kraaft, @RobUmbra and @Somethingpretentious

Stats for Season 14 Round 7:

The fastest mate was white on move 14 found in Gamelink White: mmgoo, Black: upgoerfive.
The fastest draw was found in Gamelink White: anjablaserr, Black: superbohemio.
The fastest resign was black on move 10 found in Gamelink White: babatunde1, Black: momor.
The biggest upset was 228 points in Gamelink White: fracphilo, Black: jkleebone.
The longest game ended with white on move 86 Gamelink White: jmvanaclocha, Black: mathijshuis.
172 was the highest ACPL in Gamelink White: adri89, Black: numberman768.
4 was the lowest ACPL in Gamelink White: prof_moriarty, Black: tianmu, Gamelink White: mmgoo, Black: upgoerfive.
Combined maximum ACPL was 311 in Gamelink White: adri89, Black: numberman768.
Combined minimum ACPL was 9 in Gamelink White: prof_moriarty, Black: tianmu.
The longest think was 10 minutes 50.0 seconds on move 6 in Gamelink White: karatugo, Black: seius.
The most time left was 44 minutes 4.0 seconds in Gamelink White: totallycrazybishop, Black: nice-try.
The most time spent was 68 minutes 16.0 seconds in Gamelink White: jmvanaclocha, Black: mathijshuis.

-PGNs-

Round 8:

The fastest mate was white on move 22 found in Gamelink White: tactix47, Black: petri999.
The fastest draw was found in Gamelink White: suprascud, Black: bennat123.
The fastest resign was white on move 15 found in Gamelink White: l23w45, Black: numberman768.
The biggest upset was 245 points in Gamelink White: scv-chess, Black: bward10.
The longest game ended with black on move 74 Gamelink White: skmuller, Black: alms.
164 was the highest ACPL in Gamelink White: noeom, Black: jmcginty15.
14 was the lowest ACPL in Gamelink White: slappybagowen, Black: mrbungler.
Combined maximum ACPL was 283 in Gamelink White: noeom, Black: jmcginty15.
Combined minimum ACPL was 35 in Gamelink White: darubaru, Black: stumpynoob.
The longest think was 12 minutes 6.0 seconds on move 13 in Gamelink White: suprascud, Black: bennat123.
The most time left was 42 minutes 53.0 seconds in Gamelink White: mad_shaving, Black: mqll.
The most time spent was 64 minutes 10.0 seconds in Gamelink White: skmuller, Black: alms.

-PGNs-

Stats for Season 6:

The fastest mate was black on move 15 found in Gamelink White: moiron, Black: somethingpretentious.
The fastest draw in 55 moves was found in Gamelink White: vetinari_computer, Black: sgis.
The fastest resign was black on move 11 found in Gamelink White: delpire, Black: seius.
The biggest upset was 414 points in Gamelink White: asian42, Black: seius.
The longest game ended with black on move 70 Gamelink White: mixalaki2705, Black: somethingpretentious, Gamelink White: numberman768, Black: traxxin.
187 was the highest ACPL in Gamelink White: isachess, Black: vetinari_computer.
13 was the lowest ACPL in Gamelink White: vetinari_computer, Black: mixalaki2705.
Combined maximum ACPL was 332 in Gamelink White: isachess, Black: vetinari_computer.
Combined minimum ACPL was 66 in Gamelink White: numberman768, Black: micromegas1.
The longest think was 6 minutes 36.0 seconds on move 12 in Gamelink White: moiron, Black: f1nn33.
The most time left was 14 minutes 51.0 seconds in Gamelink White: delpire, Black: seius.
The most time spent was 25 minutes 2.0 seconds in Gamelink White: numberman768, Black: traxxin.

-PGNs-

AWARDS

Accuracy King: with an average of 37 ACPL, somethingpretentious.
Fast Finisher: with an average gamelength of 47 ply/23 moves, kraaft.
Marathon Man: with an average gamelength of 86 ply/43 moves, grizzzly1000.
Biggest rating gain: with 260 rating points, grzybozbur.
Biggest rating loss: with 120 rating points, numberman768.

Chess Puzzles

by @forhavu, @kraaft and @okei

Click on the images for the solution.


Tactix47 (1828) - petri999 (1866) Gamelink. ⚫ Black to play.	SirNateALot (1713) - VladimirNotevski (1764) Gamelink. ◯ White to play.	Grizzzly1000 (2017) - Mahmut69 (1832) Gamelink. ⚫ Black to play.	phib (1533) - EmreA (1313) Gamelink. ⚫ Black to play.
Adri89 (1760) - Numberman768 (1617) Gamelink. ⚫ Black to play.	swandog (1686) - iHATEblitzANDbullet (1830) Gamelink. ⚫ Black to play.	paintman (1819) - rliu347 (1765) Gamelink. ⚫ Black to play.	EmreA (1363) - Kenovski (1623) Gamelink. ⚫ Black to play.
suprascud (1416) - Bennat123 (1337) Gamelink. ◯ White to play.

Chessboard images provided by lenik terenin.

Please feel free to join #lichessledger on slack if you would like to contribute towards the ledger in any way, or provide any feedback. Both are highly encouraged and appreciated. Thank you for reading.

Thanks to @adande1, @colwem, @forhavu, @kostasvl, @kraaft, @moistvonlipwig, @okei, @pepellou, @RobUmbra, @Somethingpretentious, @SpiteKnight and @toni2147.