Bug in Game Explorer?

Looks like there is a bug in Game Explorer.

Try this:
-    Select “Master games”, and enter the following moves: 1.d4 c5 2.Nf3 Nf6
-    At this stage, toward the bottom, you should see only 11 games in which the continuation is 3.Bg5
-    Now play 3.Bg5: out of the sudden, you have about 2000 games, rather than the 11 predicted at the previous step!

Is this some sort of bug?

Thanks,

Andrea

Well, it's what they intended it to do. I think it should have worked differently, but they didn't listen for some reason :-)

The numbers after each move is the number of times that that position has occurred, regardless of move order.

The position after 1.d4 Nf6 2.Nf3 c5 3.Bg5 has occurred 11 times (probably in various different move orders). But the position after 3...e6 has occurred 1951 times -- obviously also in different move orders.

So the numbers don't give you any information about which move is best!

It's a pity there isn't one game with 1.e4 e5 2.Nf3 d6 3.Nxe5?? in the database. If there was, it'd tell you that 3...Nf6 transposing to the Petroff was overwhelmingly the most common choice, and besides somebody also played 3...dxe5 once :-)

Thanks Scarblac. I am not sure I understand the rationale behind, but I guess I will live with it... However to me this makes the whole explorer not that usable, if you compare it with chessbase for example!

Thanks a lot anyway for having taken the time to reply to my post!

Andrea

Unfortunately guys, this is the way all opening explorers work. There is no way possible to do what you're suggesting, which is essentially to look one move into the future to determine how many games MIGHT occur in the position, because its a neverending problem of recursion. You would have to look infinitely into the future to find all games that might possibly transpose back into the position you are looking at. The game explorer is based on POSITION, and not MOVE ORDER, which is much more helpful, because its more helpful to know the outcomes of a given position regardless of move order, than it is know how only the outcomes of games that had the EXACT move order you entered.

the problem you are describing about 11 games and then seeing 2000+ on the following page is because of move order transpositions, and there is no way to solve for that, if you could, than we could essentially solve the game of chess. :)

it seems to me that FEN notation could do this, i think, because there are only a few ways to manipulate, you could transform everything into FEN and then explore through that direction. Instead of recognizing what happens next on a PGN map, you could look have each position being static, aka not showing how often something happens next, but how often a certain position occurs, and just have the moves edit an FEN board

I usually use Scid's Tree Window, which doesn't work this way, and say 365chess.com's doesn't either -- it correctly says that after 1.e4 e5 2.Bb5 Nc6, 3.Nf3 occurred 23 times in its database, and then when you go to that position, it has tens of thousands of moves again. In contrast, chess.com says that 3.Nf3 occurred 66,360 times of 11.

The difference is that chess.com works only with positions, it says "move x is possible here, and the position after it occurred so many times", while other explorers look at both moves and position -- "in this position, this move occurred so many times".

I don't know what you mean with the bit about what MIGHT happen, etc. I mean, even the previous version of chess.com's explorer did this correctly.

I (very respectfully) disagree - that's not the way, for example, ChessBase works, just to name one...

Showing stats by position is much more helpful than showing stats by exact move order. You're chances of winning the game:

1. Nf3 Nf6 2. Nc3

are EXACTLY the same as your chances of winning the game

1. Nc3 Nf6 2. Nf3

So it wouldnt make sense to show white wins/black wins/draws based on exact move order when it doesnt actually affect the ultimate outcomes of the games.

"In contrast, chess.com says that 3.Nf3 occurred 66,360 times of 11."

You are simply misunderstanding it. Thats not at all what it's saying, its saying that the positions reached after X move, occurred Y number of times, and these are the results of all of those games. So after you play 3. Nf3, that position has been reached in 66,360 times and its showing you the % outcomes.

If it were to show exact move order, then it should misleadingly tell you that Maybe white won ALL of those 5 games with the EXACT move order of 2. Bb5, which would misleadingly tell you that you have a 100% chance of winning if you play that line, which is entirely false, when you look at all 66,360 games that have that exact position.

As I stated in my previous post, it is much more helpful from a stats point of view NOT to look at exact move order because the outcome of the game is not affected by exact move order.

But by that logic shouldn't the stats be for the position that would result, not the position that you're currently looking at?

In your example, when you're looking at the opening explorer from the second move in the first variation I'd expect the stats for Nc3 to be identical to the stats for Nf3 when looking at the opening explorer from move two of the second variation.

In it's current implementation, the stats are restricted to only those games that went from the current position to the resulting position.

@Jay: Good point about the stats win/draw/loss, I hadn't thought of that. They are better this way.

But that's just one part of what it shows, the other is showing which moves are popular in a given position. Or at least, that's what people think it shows, but it doesn't. For instance 1.Nf3 e5?. It was played six times, has ten different second moves, with 2.e4 overwhelmingly popular. I'm sure in reality 2.Nxe5 was played in almost all cases... I know what causes this, it's not move frequencies, just frequency of the resulting position. But that's just totally confusing...

So is it possible to keep the win/draw/loss stats as it is now, and the move frequencies as it used to be? :-)

sorry, but what you say is just wrong. 1. Nf3 Nf6 and then game explorer shows 35 games with 2. Nc3

same exact thing it shows if you do 1. Nc3 Nf6 and then the game explorer shows 35 games with 2. Nf3

same stats for same position

Apologies -- my mistake.  This used to be an issue, I didn't realize it had been fixed.

Wait a moment though, I've just played through the original poster's example and it's exhibiting exactly the problem I've described.

The number of moves and overall stats should be for the resulting position irrespective of what position preceded it (i.e. the one currently being viewed).  There are more than 11 games in the database that have the same fen as the position that results after 3. Bg5 so I'm unclear on why it only shows the count and stats for the 11 games that presumably arrived at the 3. Bg5 position from the position currently being viewed.

Well, as simple, easy to understand example:

1 e3 e6 (8 games)

2 e4 ... (2 games)

2... e5 (128000+ games)

[edit: and strangely, there is only 1 game where this occurs after 2 moves, all the rest are after 1 move, so not even the number of moves are important]

Exactly what I meant with my original post! That's why I still don't understand the replies saying that all is well and good because transpositions are taken into account...while to me it looks like they are not, at least in the stats showed after each move. Which is kind of key, because you cannot really click on every single move of every single line to see what the real situation is, no?

There's definitely a break here.

My feeling is that the counts and stats should be based on the query of all positions in the database with the same fen as the position that would result from the move being made.

That the position currently being viewed is really not relevant beyond restricting the list of fens searched to only those positions that can legally be arrived from it in one move.  Note that there doesn't even have to be a game where the move was played from the displayed position for a result to be relevant so all legal moves should be considered.

I am perfectly fine with the DB system, but I did find it funny the other day I was exploring some games and found a weird en passant position, the DB was showing a list of possible moves one of which was en passant, not a legal move for the sequence I put in :)

Interesting -- the position in which the en passant was possible would have had a different fen from the one that you were observing (in which it is not) so it should be relatively easy to work out whether the two are equivalent or not. 

Sounds like the possibility of en passant (and possibly castling rights, although far less likley to occur) are ignored when determining wether two positions are the same.

I'm going to answer your posts one post at a time, but I'm running out of time and patience...you guys are all just confused and mistaken. :)  here goes....