|
|
|
January 5th, 2010
02:38 pm - e-mail from Professor Glickman
my livejournal is now defunct, but I wanted to post this info somewhere where a google search would find it, which is why I am posting one more time.
the glicko is the new chess rating which allows you to carry forward errors with a rating (someone who hasn't played in 20 years has a less reliable rating than someone who plays all the time). (see http://math.bu.edu/people/mg/glicko/glicko.doc/glicko.html)
I was playing Chess on FICS today and I got into a terrible argument about your rating system. Everyone there kept on repeating the fact that it is impossible to use a formula to transfer a rating from glicko to elo (or elo to glicko).
My feeling after reading your explanation of the glicko rating system (I have read Arpad's book long ago), is that it is not only possible to create a formula (at the very least for a single point in time) to transfer ratings from elo and glicko and vice versa, but that you could also calculate the possible errors involved in such a translation.
At FICS everyone said it is "impossible" because the "errors are too large."
What is your opinion on this?
--- Eric
Hi Eric,
I don't see a huge problem determining an empirically-based formula that translates a Glicko rating to a mean Elo rating using standard statistical methods on a set of players that have both Glicko and Elo ratings. I fairly regularly carry out a similar translation from FIDE ratings to USCF ratings using such methods. While it's true that it's impossible to translate a Glicko rating uniquely to an Elo rating, the task that is fairly straightforward is to translate a Glicko rating to a *mean* Elo rating, that is, among all players with a particular Glicko rating (or near that rating) one can determine the approximate mean Elo rating. The mean Elo rating can be used as an approximation for the translation. One could also always report the standard deviation around that mean Elo rating, which would give information about how inexact the translation is.
- Mark
Mark Glickman is like Arpard Elo reincarnated. It feels weird to be able to e-mail him and ask him questions.
|
November 19th, 2009
12:00 am
|
December 7th, 2008
01:00 pm - crafty and the nokia
this is why I love the nokia. Open source unix, you can run anything if you go to the trouble of compiling it. Anyway, tvijlbrief@gmail.com has compiled crafty, a super strong chess program for the nokia. It's not the strongest chess program in the world, the rybka is, but crafty unlike rybka is free, and crafty will give a strong game even for a GM if you have a fast enough computer. Not so strong on a little nokia but enough to beat most wood pushers. By default the nokia comes with crappy gnuchess. Crafty on the other hand, is an excellent program. Here is this link for the download:
http://www.v7f.eu/public/n800/chess/
Download the appropriate links.
N800/N810 http://www.v7f.eu/public/n810/crafty_20.9_armel.deb, and http://www.v7f.eu/public/n810/eboard_1.0.3-9_armel.deb
N770: http://www.v7f.eu/public/n770/eboard-N770_1.0.3-9_armel.deb and http://www.v7f.eu/public/n800/chess/crafty_20.9_armel.deb
If you don't have it already make sure to install some xterm program, I use osso-xterm.
Then you will probably need to install crafty by hand by gaining root because the libraries will probably not match up exactly with the build that tvijlbrief was using so . . .
put
N800: http://eko.one.pl/maemo bora user
770: http://eko.one.pl/maemo mistral user
into your app manager, then install becomeroot, so now you can do
sudo gainroot
with a root shell, install the downloads with
dpkg -i --force- xxxxxxxx.deb
. . . for the appropriate files, the force option will let it use old libraries. After it is installed, download your pgn libraries. I have been using the regular start.pgn to choose the opening moves, and the crafty 2006 game files for my opening book :-) It's fun to customize your opening book.
ftp://ftp.cis.uab.edu/pub/hyatt/pgn/start.pgn
ftp://ftp.cis.uab.edu/pub/hyatt/crafty.pgn/Crafty06.zip
Extract the pgn files as needed and then add them to your library. I have kept the default locations.
To add the libraries start up crafty and use the book command
cd MyDocs/.games/EBoard
crafty
book create Crafty06.pgn 60
books create start.pgn 60
Now you have crafty with some crafty ICS games for its opening. Strong!
Thanks tvijlbrief@gmail.com for compiling crafty. I think stuff like this is why the Nokia is so much fun. You get to download open source programs and configure them however you like. . . and don't get nickled and dimed by the manufacturer. Oh and thanks to Dr. Robert Hyatt for creating the Crafty program in his spare time over the last 20 years, and allowing us patzers to download it for free.
|
July 20th, 2008
September 28th, 2007
05:51 pm - 1 bit bufferred/latched i/o domino computer style
The signal knocks over a domino for the latch/bufferred i/o blocking a path with the first a>b switch of this circuit. When the clock signal comes through, you need to NOT the latch. This is done by making another of those a>b gates. Oh, you have to make sure that the curved path is longer than the straight path on the right hand side of this diagram, that's how the a>b gate works.
a b c
0 0 0
0 1 0
1 0 1
1 1 0
a>b
a b c
0 0 0
0 1 1
1 0 1
1 1 0
This is the XOR gate from before.
a b c
0 0 0
1 0 1
0 1 1
1 1 1
Or gate.
|
September 27th, 2007
11:17 am - an improved domino computer, the half adder (domino computation)
so there's a few domino computers out there, here's one:
http://www.pinkandaint.com/oldhome/comp/dominoes/
his adder has quite a few problems one notable one is that it fails a lot because of this function:
He made a NAND gate which relies on a sync signal, that is someone has to flip over over dominos at the exact right moments, not a second too early or too late. Anyway it's kluzy and it has problems, and the dominos have to be exactly the right size (you can see that they have to be aligned very precisely or else it fails) besides the fact that it requires a sync, very troublesome.
here's another domino computer:
ok that one's a lot better, he actually has videos of his domino computer doing some math,
http://kybernetikos.com/2007/03/01/domino-computation/ for the article and here for the vids: http://www.youtube.com/watch?v=5d1R0zr91Ao, http://www.youtube.com/watch?v=m2ECfgmwaRg, http://www.youtube.com/watch?v=jrtxulLmisw
he uses two "b > a" gates (you make two domino lines cross paths the first one there wins, but since a has a shorter path b only wins if a is off and b is on), which are fine, but it has one glaring aesthetic flaw. and that's his AND gate:
He's got such a big klunky block there it's designed so that if both paths hit, you get something going through, but if only one hits, then nothing happens, or an AND. The b path basically kicks the legs out from under the big block so that the a path can push it over. Having that big block is ugly, and doesn't feel right how is that a domino it's a big contraption, you might as well throw in a lego machine while you're at it. Not aesthetically right.
here is a truth table for "a AND b = c"
a b c
0 0 0
0 1 0
1 0 0
1 1 1
So anyway it seems that domino computers should really play to the domino-ish style of the dominos falling that is no stupid sync lines where you have people hitting the domino at the exact right time, no big fat block in the middle just you know, lots of dominos maybe an incline or decline just like those big domino tricks you see.
First of all, everyone knows you can make any computer with a NAND gate. Here's the truth table for the NAND:
a b c
0 0 1
0 1 0
1 0 0
1 1 0
That's not going to be possible with a domino computer. 0 in dominos means the domino is standing up. So you're asking for a pile of dominos, all standing up, with nobody pushing on them, to have an output where the result is, a domino falls over. Not possible, unless you cheat somehow (like that pinkandaint computer domino guru did by adding sync signals and weird domino patterns).
So forget the NAND. one thing dominos are really good at doing is merging into one line. You have dominos in line a and dominos in line b, you merge them voila, you have performed an OR operation.
a b c
0 0 0
0 1 1
1 0 1
1 1 1
Next, is another domino pattern that seems to play well to the strengths of the dominos:
If you hit it on the left, it will go past the left part of the circle, and then strike on the right hand, going up to the top, if you it on the right, it will up the left hand side of the circle, and if you hit it on both sides at approximately the same time, they crash in the middle and do nothing. This is a really good domino logic pattern because you get all this leeway, if you make that middle section bigger, you can increase the tolerance of this "circuit", so your dominos don't have to be all super precise. Now to figure out which logic function this thing is.
a b c
0 0 0
1 0 1
0 1 1
1 1 0
Ok, so that's an XOR function. Now, if I want to make an adder, I have a sum and a carry bit, inputs, a and b,
a XOR b = s
a AND b = c
Excellent, the above circuit the XOR does my sum, now, I just need an "AND" function, and I can build an AND from just XOR and OR, like so: (a OR b) XOR (a XOR b)
a b (a OR b) (a XOR b) (a OR b) XOR (a XOR b)
0 0 0 0 0
0 1 1 1 0
1 0 1 1 0
1 1 1 0 1
and you can see that last column is in fact the AND operation for the carry bit we need.
I don't have time right now to diagram how to build this actual circuit with dominos, maybe later, but I do think that using the XOR domino gate is a real improvement in domino computer circuits over the previous work of both http://kybernetikos.com/2007/03/01/domino-computation/ and http://www.pinkandaint.com/oldhome/comp/dominoes/
i don't remember if it's possible to be turing complete with just OR and XOR gates.
|
|
|
![[info]](http://l-stat.livejournal.com/img/userinfo.gif){kind=small}
bonerici's journal