KeithG33 commited on
Commit
7854635
·
1 Parent(s): 6dd4090

run test to fill dataset info, fix bug in .py

Browse files
Files changed (2) hide show
  1. ChessBot-Dataset.py +2 -1
  2. README.md +24 -2
ChessBot-Dataset.py CHANGED
@@ -25,7 +25,8 @@ class ChessPGNDataset(GeneratorBasedBuilder):
25
  )
26
 
27
  def _split_generators(self, dl_manager: DownloadManager):
28
- from pathlib import Path, os
 
29
 
30
  if self.config.data_dir:
31
  repo_root = Path(self.config.data_dir)
 
25
  )
26
 
27
  def _split_generators(self, dl_manager: DownloadManager):
28
+ from pathlib import Path
29
+ import os
30
 
31
  if self.config.data_dir:
32
  repo_root = Path(self.config.data_dir)
README.md CHANGED
@@ -3,15 +3,37 @@ configs:
3
  - config_name: default
4
  data_files:
5
  - split: train
6
- path: "train/*.pgn"
7
  - split: test
8
- path: "test/*.pgn"
9
  license: mit
10
  tags:
11
  - code
12
  size_categories:
13
  - 100M<n<1B
14
  pretty_name: ChessBot-Dataset
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
15
  ---
16
 
17
  This repository contains a gigantic curated chess dataset meant for machine learning. Currently the dataset contains approximately **700 million positions** in **PGN format**, split across 1000 files. Huge credits to the following main sources:
 
3
  - config_name: default
4
  data_files:
5
  - split: train
6
+ path: train/*.pgn
7
  - split: test
8
+ path: test/*.pgn
9
  license: mit
10
  tags:
11
  - code
12
  size_categories:
13
  - 100M<n<1B
14
  pretty_name: ChessBot-Dataset
15
+ dataset_info:
16
+ features:
17
+ - name: state
18
+ dtype:
19
+ array2_d:
20
+ shape:
21
+ - 8
22
+ - 8
23
+ dtype: int8
24
+ - name: action
25
+ dtype: int16
26
+ - name: result
27
+ dtype: int8
28
+ splits:
29
+ - name: train
30
+ num_bytes: 65545518162
31
+ num_examples: 636364254
32
+ - name: test
33
+ num_bytes: 7298385021
34
+ num_examples: 70858107
35
+ download_size: 7621976329
36
+ dataset_size: 72843903183
37
  ---
38
 
39
  This repository contains a gigantic curated chess dataset meant for machine learning. Currently the dataset contains approximately **700 million positions** in **PGN format**, split across 1000 files. Huge credits to the following main sources: