Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
"""Connectors for datasets built from the Lichess open database.

Keep this regular package so pkgutil.walk_packages can discover its connector children.
"""
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
"""Lichess Clock-Blunder: clocks and engine evaluation of real games, and whether a blunder follows."""

from timenet_connectors.datasets.lichess.clock_blunder.connector import (
CONNECTOR as CONNECTOR,
ClockBlunderConnector as ClockBlunderConnector,
)

Large diffs are not rendered by default.

Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
# yaml-language-server: $schema=https://docs.timenet.ai/schemas/dataset-card-v1.schema.json
yaml_schema_version: 1
dataset_id: lichess/clock-blunder
dataset_version: 1.0.0
name: Lichess Clock-Blunder
description: >-
60,358 windows from 15,205 real rated Lichess games (48,055 train, 6,136 validation, 6,167
test), each holding three per-ply series: White's clock and Black's clock in seconds, and the
Stockfish evaluation in pawns from White's view, as recorded by Lichess in the game's PGN. Each
window ends just before one player's move, and the task asks whether that player blunders within
their next three moves. A blunder uses Lichess's own rule: a move that drops the player's
winning chances by 0.3 or more on the -1 to +1 scale, with winning chances 2 / (1 +
exp(-0.00368208 * centipawns)) - 1 and a forced mate counted as +1 or -1. The answer is computed
from the recorded evaluations, not read from the ?? marks of the PGN, which do not always agree
with it. The answer is yes in 13,761 windows (22.8%). The games are the complete games with an
evaluation and a clock on every ply among the first 64 MiB of the April 2017 standard rated dump
(a fixed byte range, so every build sees the same games). Lichess records no evaluation for a
checkmating move, so no kept game ends in checkmate, and 7,273 of the kept games end on time. Up
to four windows are drawn from each game at a seeded random ply, so the share of yes answers
matches real play. The split is by game, so no game appears in two splits. The rationale of each
task is written from a template that reads the window's own numbers and the answer; it is
generated supervision, not expert commentary. Player names are not kept.
license: CC0-1.0
source_url: https://database.lichess.org/
domains:
- general
tags:
- chess
- games
- human-behavior
- lichess
- clock
- classification
- cot
- generated-supervision
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
# zstandard decompresses the Lichess dump, which ships as .pgn.zst.
zstandard>=0.22

Large diffs are not rendered by default.

1 change: 1 addition & 0 deletions pyproject.toml
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ dev = [
"edfio>=0.4.14",
"wfdb>=4.1",
"xlrd>=2.0",
"zstandard>=0.22",
]
docs = [
"zensical",
Expand Down
80 changes: 75 additions & 5 deletions uv.lock

Large diffs are not rendered by default.