|
|
|
Topic: Opening classification for games
| |
|
|
Author
| Message |
|---|
|
|
Hmm. Categorizing games is very tricky. Let me explain how the system does it, then we can discuss if there are better ways.
I created an opening database that is indexed by board position, so basically, given a position I can search the database and it will give me all the openings that reach this position at some point.
Given a game to categorize, what the system does is this: starting from the end, it searches all the board positions back to the first move until it finds one position that matches the final position of any of the openings in the database.
Now your first example:
| 1. e4 c5 2. Nc3 g6 (so far a closed Sicilian) 3. Nf3 Nc6 4. d4 cxd4 (but not anymore... this is now a garden variety Sicilian) 5. Nxd4 Bg7 6. Be3 (more specifically an accelerated dragon...) |
The accelerated dragon variations I have on the database (actually called accelerated fianchetto) are all 7 moves or longer. The last position in the game matches at least one of the variants of this opening, but the match is for an intermediate position, not the final one, so the system doesn't use it yet. If you continue making the moves in this opening when you reach the end position the game will update its category.
Does this make sense? Any comments?
Miguel
|
|
|
I'm curious. Why does B20 sicilian chochrane attack end after three moves? Is that the end of the opening?
|
|
|
| Why does B20 sicilian chochrane attack end after three moves? |
The three B20 Cochrane lines that are in the ECOMast database are:
B20 - Sicilian: Cochrane (Burger) attack
1.e4 c5 2.Bc4 *
B20 - Sicilian: Cochrane attack
1.e4 c5 2.Bc4 e6 3.Nf3 Nc6 *
B20 - Sicilian: Cochrane attack
1.e4 c5 2.Bc4 Nc6 3.Nf3 e6 *
|
|
|
Since the transpositions are handled from the seventh move and on then we might be prone to have inaccuracies (that are also seen even in the Yugoslav ECO) 
My suggestion would be to let the system walk through the encyclopaedia as if it were flipping pages and let it start with A00 on the first move and let's say it works backwards:
1.e4 (C00-C99, B00-B99)
2...c5 (B20-B99)
3.Nc3 (B23-B26)
3...Nc6 (B23-B26)
4.Nf3 (And here the program recognizes that this position could have been reached by 1.e4 c5 2.Nf3 Nc6 3.Nc3 and that it is now working with the ECO code of B30 but will wait for a couple more moves)
4...g6 (This is still B30)
5.d4 cxd4 6.Nxd4 (and now this is B34)
I see now that the program started handling transpositions from the seventh move because it asigned the game a wrong opening.
I would suggest letting the program browse ECO spotting transpositions as it goes and if the opening falls with two classifications let it say it...
|
|
|
I think you still don't understand how the sytem categorizes a game. The opening database stores board positions for all the moves in all the openings. Then a board position in the database can be final or intermediate. A final board position for an opening is one that is reached after the complete line of that opening is played. The final board positions are unique, which means that there are no two openings that have the same final board position. To categorize a game the system searches the current board position only looking at the final opening positions, the intermediate positions are not searched.
| I would suggest letting the program browse ECO spotting transpositions as it goes and if the opening falls with two classifications let it say it... |
Well, this is going to be difficult. Imagine that after 1.e4 there are 1400 openings that match the board position, I can't possibly show them all! That's why only final positions are searched, which guarantees that only zero or one openings will be found.
For your information, here is a text version of Paul Onstad's ECOMast database, which I use here:
http://pgn.freeservers.com/ECOMast.txt
Take a look and if you find a specific example where an opening line from this txt file was not chosen here please let me know.
Miguel
|
| Previous 1 2 |
|
|
|
