Great request! It's already in the To-Do list :)
Game Explorer feature request
Sort:
Sep 9, 2008
0
#3
Another thing: I have the feeling the game explorer doesn't pick up all games that lead to a certain position. For example, when I use the following sequence:
1. Nf3 Nf6, 2. Nc3 Nc6, 3. e4 e5, I get slightly over 10,000 games in the database. But when I use 1. e4 e5 2. Nc3 Nc6 3. Nf3 Nf6, there are 25,000 games in it. The position is however exactly the same (ok, maybe not the best opening ever, but it's just an example) so it would be great if there would be a function to see all the games from that position, as opposed to the games which used that sequence.
DrPhil wrote:
Another thing: I have the feeling the game explorer doesn't pick up all games that lead to a certain position. For example, when I use the following sequence:
1. Nf3 Nf6, 2. Nc3 Nc6, 3. e4 e5, I get slightly over 10,000 games in the database. But when I use 1. e4 e5 2. Nc3 Nc6 3. Nf3 Nf6, there are 25,000 games in it. The position is however exactly the same (ok, maybe not the best opening ever, but it's just an example) so it would be great if there would be a function to see all the games from that position, as opposed to the games which used that sequence.
I have noticed that exact same thing, but have not had time to post it yet. If the game progresses 1. e4 e5 2. Nf3 Nc6 it recommends 3. Bb5 (Ruy Lopez), but if you get to the exact same position by 1. Nf3 Nc6 2. e4 e5 the top move on the list (with a much smaller number of games) is 3. d4 (Scotch game). You would think that both sequences arriving at the same position would produce identical numbers, but they don't.
Also there are times in which a particular move will say something like 4 games (75% white wins, 25% black wins), but when you click on that move, the top move on the list for your opponent has 20 games with only 30% win for white and 70% for black. These are hypothetical numbers of course, but I have encountered this situation many times which causes me to question how they calculate the winning percentages.
Sep 9, 2008
0
#5
JFercan wrote:
But if you arrive to the same position through different sequences, wouldn't this make the games different? Why would they be counted as the same?
Well, one of the uses of the explorer is to check what kind of moves are commonly played in the position you have in one of your own games - and if the database ignores half of the games in it (which have the same position, so there are just as relevant for this purpose) it doesn't work as it should.
It's a great tool, but it will be a lot better if this quirk is solved.
It would also be great if there was some way to step through an opening "tree". Right now, as you step through openings, the explorer only lists "related" openings when you are close to them (not sure what the heuristic is exactly). I would really love to be able to visually see a graph of all opening variations and their relationships, with the opportunity to selectively explore each.
At the moment, all the openings stand alone, and the relationships are not clear. (This is a similar complaint to the above, where openings that lead to the same position are considered distinct, even though they are actually very alike).
It's a relatively minor thing though.
Sep 9, 2008
0
#7
The percentages shown in opening explorer are correct. This has been answered in other forum threads I believe. If you want to provide an exact example of where you think the percentages are wrong, then I can explain to you why they are right. :) Every position is looked at as a discrete position, regardless of the move order that took place to get there.
Then how do you account for the same postition giving different numbers based on the order?
For example: The most common opening combo of (1.e4 e5) (2.Nf3 Nc6) gives the following numbers:
|
Move |
Games |
White |
Draw |
Black |
|
3.Bb5 |
128,105 |
41.6 |
31 |
27.4 |
|
3.Bc4 |
58,710 |
43.7 |
22 |
34.3 |
|
3.d4 |
34,803 |
45.2 |
24.8 |
30.1 |
|
3.Nc3 |
20,583 |
38.5 |
27.6 |
33.9 |
|
3.c3 |
3,346 |
41.7 |
24.2 |
34.2 |
|
3.d3 |
716 |
34.9 |
19.1 |
45.9 |
But the same position by (1.Nf3 Nc6) (2.e4 e5) yields these numbers:
|
Move |
Games |
White |
Draw |
Black |
|
3.d4 |
34,803 |
45.2 |
24.8 |
30.1 |
|
3.c3 |
3,346 |
41.7 |
24.2 |
34.2 |
|
3.Bb5 |
867 |
40.8 |
30.9 |
28.3 |
|
3.d3 |
716 |
34.9 |
19.1 |
45.9 |
|
3.Bc4 |
363 |
43 |
24.2 |
32.8 |
|
3.Nc3 |
274 |
38.7 |
30.7 |
30.7 |
In fact, if you tried an even different route: (1.Nf3 e5) (2.e4...) it will tell you that there is only ONE GAME with the position 2... Nc6, and black won it.
???
Sep 10, 2008
0
#9
So, the uniqueness of a position is defined by the fen string. The fen string includes the following variables: move number, castling, AND en passant squares. So, even though the positions LOOK the same based on the position of the pieces, they are NOT the same because the en passant squares are different. :)
Sep 10, 2008
0
#10
En passant should not matter if there is no capture to be made.
It may be true that Nf3 Nc6 e4 e5 is not the same as e4 e5 Nf3 Nc6 because of the fifty move rule, but for the sake of opening explorer, this should also not matter. Perhaps there can be an option (default it to true) "ignore 50 move rule".
But the en passant really should not matter, if there is no capture to be made.
Sep 10, 2008
0
#11
Unfortunately thats not how fen strings work. I wish that were the case, but its not. Maybe there is some way we can combine all those fen strings, but its not likely.
In other words, there is no way to account for transpositional possibilities.
Billium248 wrote:
In other words, there is no way to account for transpositional possibilities.
Use the opening explorer ar www.365chess.com. It work the way you want.
Great site! Thanks!
Sep 10, 2008
0
#15
jay wrote:
Unfortunately thats not how fen strings work. I wish that were the case, but its not. Maybe there is some way we can combine all those fen strings, but its not likely.
I would suggest exactly that. The expected user behavior is that Nf3 Nc6 e4 e5 should be the same position as e4 e5 Nf3 Nc6. Perhaps add it to the feature request list?
Sep 10, 2008
0
#16
orejano wrote:
Billium248 wrote:
In other words, there is no way to account for transpositional possibilities. Use the opening explorer ar www.365chess.com. It work the way you want.
Wrong. They have the same (or worse) problem:
http://www.365chess.com/opening.php?m=3&n=136&ms=Nc3.c5&ns=65.136
Notice how that shows that only 604 games include the move e4 from the above position. However, after you click e4, you'll notice there are thousands of games. And if you play the move order e4,c5 you'll notice that Nc3 has 60k games. We have already checked all the other major sites that have explorers and all of them have these same pitfalls, or worse pitfalls. :)
Sep 10, 2008
0
#17
ozzie_c_cobblepot wrote:
jay wrote:
Unfortunately thats not how fen strings work. I wish that were the case, but its not. Maybe there is some way we can combine all those fen strings, but its not likely. I would suggest exactly that. The expected user behavior is that Nf3 Nc6 e4 e5 should be the same position as e4 e5 Nf3 Nc6. Perhaps add it to the feature request list?
Maybe. Its a very large project to rescript the calculation of 4 million plus games' fen strings. We're talking about weeks just for the script to run. Thats the problem of working with these massive data sets, it becomes very costly to try and make architectural changes such as this.
Sep 10, 2008
0
#18
Weeks isn't so bad, is it? The script shouldn't require site downtime, and once you figure out the user expected behavior it shouldn't be too difficult to write. As a user of the system, I just don't expect that the FEN notation of two positions needs to be identical in order for me to consider them identical.
Sep 10, 2008
0
#19
Yeah, I agree with you. I'm currently working on some other features/fixes, so when we've got time and if it seems doable, then I can see if I can script a fix for this.
Sep 10, 2008
0
#20
ozzie_c_cobblepot wrote: Weeks isn't so bad, is it? The script shouldn't require site downtime, and once you figure out the user expected behavior it shouldn't be too difficult to write. As a user of the system, I just don't expect that the FEN notation of two positions needs to be identical in order for me to consider them identical.
jay wrote: Yeah, I agree with you. I'm currently working on some other features/fixes, so when we've got time and if it seems doable, then I can see if I can script a fix for this.
Have a care that you don't accidentally 'solve chess' if you make this change.
I would love for a way to enter an FEN number into the Game Explorer to immediately call up a board position, rather than having to step through it each time to get it to where I want it to be....