Datasets:
license: cc0-1.0
task_categories:
- text-generation
language:
- en
pretty_name: Annotated chess games and studies
size_categories:
- 1K<n<10K
Dataset Card for Dataset Name
Dataset Description
- Point of Contact: [email protected]
Dataset Summary
This dataset consists of annotated chess games and chess studies by humans. It has two subsets, the first one "lichess" consists of the top lichess studies scrapped from lichess.org. The "others" subset mainly consist of games from https://www.angelfire.com/games3/smartbridge/
Supported Tasks and Leaderboards
It is intended from training chess text generative models.
Languages
The main language represented is english, although some other languages might be present in unsignificant amounts.
Dataset Structure
How to use:
from datasets import load_dataset
import chess.pgn
import io
dataset = load_dataset("Icannos/chess_studies", "lichess", streaming=True)
for d in dataset['train']:
pgn = io.StringIO(d['text'])
game = chess.pgn.read_game(pgn)
print(game)
break
Data Instances
Example of annotated game / study from lichess. The annotations includes arrows and circles drawn on the board in addition to natural language commentaries and sometimes computer evaluation.
[Event "🇷🇺 Petrov Defense 🇷🇺: Nimzowitsch Attack"]
[Site "https://lichess.org/study/OnPMlzHT/oG7xbZFE"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[Annotator "https://lichess.org/@/LeninPerez"]
[ECO "C42"]
[Opening "Russian Game: Nimzowitsch Attack"]
[UTCDate "2021.02.11"]
[UTCTime "00:54:33"]
[Variant "Standard"]
1. e4 { Do you remember the movements from the previous chapter? I hope so, because you should do them now :D } 1... e5 { That's! } 2. Nf3 { And now? } 2... Nf6 { Great job! } 3. Nxe5 { You will find this frequently in your games with this defense. That is, the most common in move 3 is that the white player takes the pawn.
How can you drive the white knight of e5? } 3... d6 { Very well! [%csl Re5][%cal Rd6e5] } 4. Nf3 { You know what you have to do now, right? } 4... Nxe4 { Excellent, you get the pawn back!
The blue arrows represent all the options the white player has to play now. [%cal Bd2d3,Bd3d4,Bb1c3,Bd1e2] } 5. Nc3 { This is the Nimzowitsch Attack!
Change the knights [%cal Re4c3,Rc3e4] } 5... Nxc3 6. dxc3 { The white player must deal with the doubled pawns on the c column
Develop your bishop [%csl Gf8] } 6... Be7 7. Be3 { What would you play now?
(Psst, your king is in the center) } 7... O-O 8. Qd2 { White wants the queenside castling
Now you must take your knight to f3, what is the shortest route? [%csl Gc1,Gf6,Gb8][%cal Ge1c1] } 8... Nd7 { That's! } 9. O-O-O { This is really the Nimzowitsch Attack.
White castles long to plan a battle of attacks on opposite flanks!
Where should this knight go? [%csl Gd7] } 9... Nf6 10. Bd3 { Play 10.c5 [%csl Gc5][%cal Gc7c5] } 10... c5 { Very well! Now the white player wants to attack your king with the pawns on the queenside.
You must play as I indicate with the arrows, that is, attack the weak point a2 and improve your towers. [%csl Ra2][%cal Ba8c8,Bf8e8,Yc8e6,Yd8a5] } *
Process finished with exit code 0
Data Fields
The only field is "text". Each row contains exactly one game/pgn file in the text field.
Data Splits
A single train split.
Dataset Creation
Source Data
Lichess studies
The lichess studies consist of the first 10 pages of studies (ranked by stars) on lichess (https://lichess.org/study/all/popular).
Others studies
I relied mainly on the compilation built over the years on https://www.angelfire.com/games3/smartbridge/ and which consists of top player games.
Other Known Limitations
The annotations are mainly in english (although some are annotated in french).
Citation information
TO COME.