GameKnot: GameKnot related, Post-game computer analysis

« TAP TO LOG IN

GameKnot related: Post-game computer analysis

« Back to forum

Pages:

From

Message

Gameknot.com
18-Apr-10, 21:56

Post-game computer analysis

Announcing the newest GameKnot feature: post-game computer analysis. You can now submit any of your past games to be analyzed on the server. Each move will be evaluated and assigned a score, so you can see the progression of who was winning or losing in the game. Also, if a mistake/blunder is found, the best alternative move will be suggested.

You can request the post-game analysis via "analyze the board" page -- the corresponding link is below the board. Only for completed games of course. Up to 10 games can be requested to be analyzed per month by premium subscribers, and just 1 game per month for free members. We will likely allow more games in the future -- we just don't want to overload the servers if it proves to be very popular.

Comments? Suggestions?

chess_avenger
18-Apr-10, 22:13

excellent idea

excellent idea and a nice feature to this great site. im trying it out at the moment.

baronderkilt
18-Apr-10, 22:18

Intriguing ...

but what a blow to us opening specialists if everyone starts following the GK line~!? Hmmm... Time to become Endgame Specialist ?!

chess_avenger
18-Apr-10, 22:37

continuation to my previous post

now i have tried it looks great...

i may not be right but one suggestion.

instead of clicking individually on the various suggested moves under computer analysis.
it would be nice to show the pieces sliding to the next position.
as far as i know each new click goes to a newer position. the above would make it better.

chess_avenger
18-Apr-10, 22:50

...

Example of what I mean:

Inaccuracy: +0.81 » 7. Kd1 c5 8. Nf3 Nc6 9. Nc3 Nxc3+ 10. dxc3 Ne5 11. Qe4 Nxf3 12. Qxf3 Be6 13. Bd3 O-O-O 14. Re1 Re8
Best: -0.15 » 7. d3 Nf6 8. Nc3 Bd7 9. Nf3 Nc6 10. Bf4 Qxe2+ 11. Bxe2 Be7 12. O-O-O O-O 13. Rhe1 Rfe8 14. Ne4 Nd5

The above example shows the suggested moves. When I click on the best moves "7. d3" it shows a number "Best: -0.15 and 'Return to main line'.

Instead of this when I make click the move 7. d3 it would be better to show it visually on the board.

Hope that makes sense.

Gameknot.com
18-Apr-10, 23:16

chess_avenger, when you click on any of the analyzed/suggested moves below the board, it puts the whole line into the main list of moves on the right side, so that you can use the normal means of navigation through the moves (i.e. by using the mouse wheel, or the [<] and [>] buttons below the board, etc.). Hopefully this is what you are looking for?

chess_avenger
18-Apr-10, 23:22

thanks once again

Thanks once again. I didn't realize I had to press the left/right keys to move the pieces.

mattchess
19-Apr-10, 01:07

Excellent feature!

black_cat_hamlet
19-Apr-10, 03:38

Brilliant idea!

I'll look into it when I have got an interesting game... actually I already have one in mind...

dugite
19-Apr-10, 03:46

Good idea!Do you use your own engine for analysis?

kingdawar
19-Apr-10, 04:06

Deleted by kingdawar on 19-Apr-10, 04:37.

scarper
20-Apr-10, 06:36

I like it!

Thanks for a very interesting new feature. Two questions:

1) Will the computer analysis for a game remain accessible from the the game's board in "Past games history?

2) What is the difference between an "inaccuracy", "mistake" and "blunder"?

tugger
20-Apr-10, 08:30

Brilliant idea. I've just requested my first analysis. I am curious though, just how strong is the computer?

Scarper, I would imagine that an "inaccuracy" is the most common form of chess move, a good, but not perfect, move. A "mistake" is a move that loses a tempo when you could've gained a tempo, and a "blunder" is when material or position is lost needlessly. I'm sure there will be plenty of controversy about which category a particular move falls in, for example, if you play 27. Be4, which wins a pawn in two, but you could've played 27. Qh5!, which mates in seven, is that an inaccuracy, or a blunder? I predict the computer would call it an inaccuracy, though if the mate was reasonably simple, ie the next six moves are all checks and forced replies, then for someone of my rating, I would say it was a blunder. I think there will always be some debate here, one thing a computer cannot do is tell us how difficult it would be for someone of a given rating to find the better moves, which might affect whether a move is an inaccuracy or blunder. I think it is relative to your rating.

ironbutterfly
20-Apr-10, 09:38

what computer?

Definitely a great feature!

This is related to dugite's question above - what computer is being used
for the analysis?

tugger
20-Apr-10, 10:01

Hmm... I'm not sure my definitions of mistakes and inaccuracies are correct. In the game in requested analysis for, I sacrificed my rook for what I considered to be an easy endgame position. It told me it was a mistake, and showed me the better line. Now, while the "best" line did indeed look good, it looked more complex than the endgame I left myself with. Throughout the actual endgame, I was always up more points than the pawn I held as material advantage. Sometimes it's ok to make a "bad" move if it simplifies the position, but unless the computer detects a forced mate in the process, I don't think it takes simplification into account. I suppose the desire to make things simpler is a human trait.

Still, I'm very impressed with this, and it can only help us to improve. I was surprised to see the effect early positional moves had, for example, an early knight move of his chaged the score from +0.18 in his favour to +0.27 in my favour, boosted again by my next move. This will help those us without a good computer already to fine tune our openings, looking for that tiny advantage in the early stages, and it will also assist with endgame study, seeing how pawn moves and king moves dramatically change the score.

I intend to have your computer working hard... shame we can only request one at a time haha!

maca
20-Apr-10, 11:26

Good feature, but...

To me, it appears that the scores this system gives are rather large in magnitudes, and quite dynamic compared to other chess engines that I've used. That said, I've only used to analyze one game so far. I'd really like to know which engine they're using.

Tugger, I feel I should also warn you about following the score blindly, particularly in the openings. It appears that the opening book of this engine may not be too sophisticated, and computers tend not to do well in the early stages of openings if they're unable to consult their book. Plus, a change from +0.18 of +0.27 is microscopic; it's so small that it is probably impossible to comprehend for most players.

Regards,
MaCa.

Gameknot.com
20-Apr-10, 12:55

dugite, yes, the program that does the analysis was written in-house. It is fairly simple, but uses all the modern features common in other chess engines, such as bitboard data structures, alpha-beta pruning, killer move heuristic, null-moves, etc.

scarper, yes, any time you access the "analyze" page for a game, the server automatically includes the computer analysis if it is available (including other players' games).

Inaccuracy = 0.5-1 points difference between your move evaluation and the best move evaluation (e.g. if you lose a pawn or less), mistake = 1-3 points difference (a couple of pawns), blunder = everything over 3 points (knight/bishop and above).

tugger, we are actually not sure how to evaluate the current strength of the engine. I would guess it's at least above 2000, since it was finding inaccuracies in higher rated players' games when we were testing it. We'll probably make it play a few games against other chess engines with known strengths at some point, but it'll require more work and it's just not a priority at this point.

tugger
20-Apr-10, 13:40

maca, it went from +0.18 in his favour to +0.27 in my favour, so it actually went from -0.18 to +0.27, which I considered to be significant for a move that was not obviously bad. But thanks for the warning, I guess it's easy for a computer to analyse an endgame entirely, but it obviously can't do this with an opening. I'll exercise caution when assessing openings.

Thanks GK, I would have to agree it's over 2000, I would imagine significantly, based on the strength of opponents at that rating I have faced. It's definitely stronger than I am, that's for sure. Hopefully one day I can say that I am better than it! Thanks again for this great addition.

tugger
20-Apr-10, 14:28

Oh, a suggestion... would it be possible to have a link to a list of computer analysed games somewhere, perhaps next to annotated games at the bottom? Only once the private message has gone, one has lost a direct link, and then one has to find the analysed game amongst all the other games that haven't been analysed in the past games list.

Another suggestion... is it possible to play unrated games against the computer?

In any event, it should have a name. We can't just call it the computer. Any suggestions? I propose Gaz, for Gameknot's Analyzer.

omus
20-Apr-10, 14:49

many thanks

thanks for adding the feature. brilliant idea.

kiwisouth
21-Apr-10, 01:36

Post-game computer analysis

Brilliant - I've been looking for a method/person to do post game analysis and find GAZ (Great Name!) a great teacher. I've gone through one game and spent more time on that than on my games tonight - well done!

hektorx
21-Apr-10, 03:27

I've made a computer analysis for my last finished game..

Each move was be evaluated on the analysis, nice but I couldn't see any the best alternative move.

kreifi
21-Apr-10, 04:11

Wow

Just what I have wished for long time. Have not yet tried this but thank you so much. Best chess site. <3

levellerlevvie
21-Apr-10, 07:16

Past Games: Visible which games are analysed

In the Past Games Page (and other pages with similar info) ... Wouldn't it be useful to indicate which games are already analysed and which not.

I myself haven't asked for an analysis yet but apparently one of my recent opponents did with one of our games. I was fortunate to find out though since I only stumbled upon the analysis while browsing my last games played.

Possibly you guys have this enhancement already in the pipeline and are improving this new feature while the core is already operational but if you haven't ... please consider adding it.

Thanks!

hektorx
21-Apr-10, 07:38

Sorry!...I revised the computer analysis again and I noticed for the best moves suggestion...

But I think 'mistake' or 'best move' indication should be colorfull to be noticed easly.

Thank you.

coopershawk
21-Apr-10, 08:12

Thanks Gameknot

Nice additional and unique feature for the site.

easy19
21-Apr-10, 12:39

I do not expect that the analyze engine is able to see that the best calculated move to play is not always the best move to make.

A good example with what i mean is from the movie I robot
where the police-agent is saved from a sinking car, by a robot instead that the robot decided to safe the child. every human would decide to safe the child first a computer will safe the police agent first..

baronderkilt
21-Apr-10, 13:00

What they don't tell you ...

but I've heard a rumor (I always hear them even if I have to say them outloud ... ) that the GK box has a super secret added Human-Heuristic ... if things get too tough, a light flashes to call Tag1153 to the board to annotate 98 of a hunded trouble spots, after which a great foghorn calls Cyrano in ...??!

kiwisouth
24-Apr-10, 17:14

Analysed games listed same as annotated games

Icing on the top of this great feature would be to have a link to analysed games - same as with annotated games. cf Tugger

tag1153
24-Apr-10, 19:49

lol

hey Craig - you nailed it. I'm really a 2600+ silicon beast with a flair for storytelling

Pages:

« Back to forum