yesterday I had 1810, now I dropped back to less than 1700… within a day without forcing it.

(it is easy to achieve if you postponing losing matches and speed up winning ones, but I didn't do any of that.)

At least I got above 1800 for once…

Congratulation Hjallti, I did it for the third time last week, and I reached my peak with 1826, but I dropped back so fast too. I never manipulate my rate, and it is much more fun to do it without manipulation. I just want to ask players over 1800, how do you manage to stay over 1800 for long time? I never keep it for even a week :)

Just keep on winning. ;)

According to my graph it was the 3th or 4th time I got there.

It has some relation to stock market engineering: You have
to resign your lost games in moments when it does not cost
you much …

Ingo

Many players who want to keep their >1800 points are not playing rated tournaments/games anymore with a low number of single games because of the randomness of the game and the low k-factor in the (ELO-) points calculation. Especially the 1-point-tournaments and also the team tournaments make the most fun for me but cost really too many points…

I have an internal database of all EWN players in which I'm assigning the (ELO-) points with a much lower (adjustable) k-factor to the date when the game started and not when it ends. So, you're having also peaks in the curve but at the starting time.
These peaks are less controllable by the players because they cannot know in advance if their games will be lost or won. In my eyes, that gives you the better idea of the real (current) strength of a player.
In addition, the players are not able to 'park' themselves with a high number of points on top when they stop playing totally. So, for example, 'Jonny' as well as 'opmp' are the same person while 'Jonny' is mainly playing tournaments/games with a higher number of single games (up to 50-points-matches) and opmp the 1- or 3-point-matches. The rating of both is totally different here on LG while they are having exactly the same points in my calculation. I'm also taking all games (rated and unrated) into account as well as the score (single games) not just the game result (+/- 2). So, I think that's more realistic because I'm calculating with more information.

If you want to take a look on it you can find it on this webpage: http://bloederblitz.jimdo.com/ewn/
There's a download link below, the name is LG_EWN_DATABASE…zip. You must extract the file in a separate folder and start the exe-file. It needs 1-2 minutes to open because all values are not saved, the stats must be calculated first.
All buttons on that form are disabled, the only things you need are the scrollbars. You can choose a specific player/account with the upper scrollbar and the first curve you see is the ELO-calculation. Afterwards, you can find all opponents of the first player in the scrollbar below. By choosing a specific opponent you can see also the history of the score ratio over all games against each other. It could be that the window is much too large for your screen. I'ts optimized for a resolution of 1600 x 900.
You can find also an overview of all players in the included excel-sheet. It's german but should be understandable. Don't take the rating too serious..

I did some analysis of the LG EWN rating system a few months ago – why it is horribly broken and how to fix it – but never posted a write-up.

I agree that every game should count, not just the match outcome (that's just a gross loss of relevant information), and that the k-factor should be much lower (like backgammon).

There are a variety of ways of getting more meaningful information out of the data. Is the database of results available? What format is it in?

The child is automatically collecting the data from LG. This is an offline database. You can find the results uncoded as well as uncompressed in a txt format in that zip file.
The players are in 'SpielerDatenbank.txt',
all finished games are in 'SpieleDatenbank-finished.txt',
'SpieleDatenbank-open.txt' includes the running games (at the date of the last update).

In the window opened by the .exe file, there's a maximum width and some of the information is to the right of that. There are no scoll bars, either veritcally or horizontally, that apply to the whole window.

hmm, try to set your screen resolution to 1600 x 900. The scrollbars should be visible at the upper border of the window.

I've inserted also the win ratios of all players as a function of opponent's ELO points at the upper right corner. That's interesting for players with a higher number of games. You can observe also a learning effect because all games are taken into account as well as the average of the last 10 (adjustable) games against opponents with a specific ELO.
Take a look at 'luiser', 'Ray Garrison', 'Jean Hebert' or 'Telestes', they are able to play nearly 50:50 against the best players on this site. That's incredible. The best players are represented by the last two green bars and typically bots (with only a few exceptions)…

1366 x 768 is as high as my laptop goes. That's what I keep it on.

I got up to 1830 in the official ratings for a while. I was very briefly the top truly active human. I wonder if you think I was ever that good.

Have you tried to move the window such that you can maximize it? Normally there should be scroll bars available in full screen mode.

Calculating with a k-factor of 3 makes the ELO curve less alternating, but you need more games to reach the points of the best players and your <150 games are not much. I haven't implemeted yet what Ingo has proposed - to increase the k-factor for starters, to push their points in the first games - I need to know: For how many games, how high? That's why it is very difficult to reach a lot of points in a rating with a lower k-factor with such a few number of games (That's why I'm calculating with all single games - that are nearly 700 in your case - but still a lower number for this game).

Nonetheless, your maximim points in that system with k=3 are 1585, what represent the points to reach the current TOP 25. That's definitely good, but their maximum points are ranging much higher.
OneStone_c got the maximum peak ever with 1697. It could be that some humans would have reached a higher peak if they had played more games (eg Telestes, Theo van der Storm, Kitaktus, MarleysGhost, wurscht, Ray Garrison, Jean Hebert), but the current maximum peak of a human ever is 1657 by luiser - 73 points higher than yours, that's pretty much in that system…
You have to play more games, maybe you can beat her !? ;)

You are winning 70% of all single games against beginners, that's very strong, and you are currently in the top 50 (including all bot and inactive accounts). Here are extremely good people around, I think you can be proud of that.

Thanks. I'm also doing well at mastermoves.eu, so I figure I'm doing more right than wrong.

One approach would be to run a simulation where the same results or rted over and over, until they produce a negligible change. It makes sense to seed each player with some portion of par results, so that a few lucky or unlucky games to start don't produce silly results. Also, if you have only 2 players who play an even match or some other pathological result sets for many players, you have to award only a large fraction (say, 80%) of the points that would be awarded from the results in order to avoid undamped oscilation. I' ve taught myself a few programming languages implementing that method for American Football and chess.

The window shows no scroll bars when maximized (and when not). I suspect it's trying to draw them, but outside the visible region. That's not how I would think a Windows application would behave.

That sounds strange, I know that the program is working on other PC's. Do you have installed Microsoft Silverlight?

Your simulation sounds plausible in the first moment, but I don't know if it would produce an accurate rating that would really represent the current player strength.
If two players would have the same playing strength but the first one would have much more games or a steeper ELO curve at the beginning wouldn't he get also the higher rating position?

Hi,
Just letting you know I have an issue with your soft. I guess it is related to “the win ratios of all players as a function of opponent’s ELO points at the upper right corner”, because I didn't had the issue before and the corner remains white.

Couldn't cast string “3.1415” to Double, FormatException.

System.InvalidCastException: La conversion de la chaîne “3.1415” en type 'Double' n'est pas valide. —> System.FormatException: Le format de la chaîne d'entrée est incorrect.
à Microsoft.VisualBasic.CompilerServices.Conversions.ParseDouble(String Value, NumberFormatInfo NumberFormat)
à Microsoft.VisualBasic.CompilerServices.Conversions.ToDouble(String Value, NumberFormatInfo NumberFormat)
--- Fin de la trace de la pile d'exception interne —
à Microsoft.VisualBasic.CompilerServices.Conversions.ToDouble(String Value, NumberFormatInfo NumberFormat)
à WindowsApplication1.Naive_Child_c.WINvsELO(Int32 spielerindex) dans C:\Users\Mark\Documents\Visual Studio 2010\Projects\LG-Datenbank aufbauen_V5\Naive_child_c\Naive_child.vb:ligne 2476
à WindowsApplication1.Naive_Child_c.ComboBox1_SelectedIndexChanged(Object sender, EventArgs e) dans C:\Users\Mark\Documents\Visual Studio 2010\Projects\LG-Datenbank aufbauen_V5\Naive_child_c\Naive_child.vb:ligne 3433
à System.Windows.Forms.ComboBox.OnSelectedIndexChanged(EventArgs e)
à System.Windows.Forms.ComboBox.WmReflectCommand(Message& m)
à System.Windows.Forms.ComboBox.WndProc(Message& m)
à System.Windows.Forms.Control.ControlNativeWindow.OnMessage(Message& m)
à System.Windows.Forms.Control.ControlNativeWindow.WndProc(Message& m)
à System.Windows.Forms.NativeWindow.Callback(IntPtr hWnd, Int32 msg, IntPtr wparam, IntPtr lparam)

hmm, I think, this is the difference between a software engineer and a hobby-programmer ;)
Maybe that could come up because of the different international decimal point systems ( , or .) Do you see the ELO curve below after choosing a player?
Could you try out what would happen if you would delete the decimal places or to type in simply '3' for the k-factor. Please, start a '(New) ELO calculation, 3 should be always accepted as a number…if the window would be visible, the decimal point must be the problem, but normally the calculation of the ELO curve should also not work. That would be strange.

Curious, does the “3.1415” string has anything to do with Pi?

Yes, it was Ingo's idea/joke, but I think this factor is still a bit too high. The ELO curves are still fluctuating too much. The rating looks only steady after averaging several games.

Yes, the simulation I suggest assumes a constant strength, which would be a flaw here. If you had it already written, I think it would be worth using, but not in raw form.

I have Silverlight. I'm running Windows 7.

It is definitely an issue with the decimal point system. I didn't have enouh time to try to only use 3 (tonight maybe), but chacpi_listenanging my localization parameters to use . instead of , is enough to fix it.

The main ELO curve was visible.

Hm… “chacpi_listenanging” is “changing”. Sorry for the mess. :p

Ok, I'll change this Pi into 3 in the future to avoid those problem, but that's pretty curious, because the win ratios as a function of opponent's ELO points are accessing the values only that are shown in the ELO curve below. If this graph is shown, these values are obviously calculated correctly.

But that Dvd Avins isn't able to scroll within the window is still mysterious. That's something I probably can't change.

I know I'll have a whole bunch of losses in the future (but maybe months down the line, many are 50 point games) so can't stay there for any length of time - but I'm now at 1906!

I keep popping over 1800 and falling back down, but my current rating is 1691! Just can't resist all the 1-point tournaments.

I think the reason I've crept upwards is becase I avoid 1-pointers, and also prefer (and think I'm stronger at) backward capture.

Indeed. If I'm using the 'backward capture' results only to calculate the ELO points (with k=3) you're the strongest player of this variant.

who is the strongest in black hole, YHW?

On position #1 at black hole is OneStone_c, followed by Looser, ypercube, Hjallti, kpato, Carroll, FatPhil, kfiecio and RoRoRo.
But note, the most players have played just a few number of games of this variant. So, the points of the most players are still rising (with different steepness). That's why the current ELO points are not very meaningful.
However, you can calculate it yourself by downloading the program described above. After starting the program you have to disable the normal variants (the checkmarks on the left side: 1-point-game, 3-points-match and so on) and choose the 'black hole' checkbox only. After clicking on '(New) ELO calculation', the ranking is shown in the window at the lower right corner.
If you want to see your ELO-curve you have to find your nick-name at the scrollbar on the upper border of the window. Afterwards, you can find all results against all opponents (against which you have ever played this variant) at the scrollbar below. After choosing a specific opponent the trend of your win ratio against this opponent is plotted below. But this is usually only of interest after playing a higher number of games.

stupidest game ever.

There is a follow-up in the thread:

http://www.littlegolem.net/jsp/forum/topic2.jsp?forum=140&topic=183