feat: scripts for automated chess tournaments and analysis

contrib/fast-chess-branch.sh

@@ -0,0 +1,82 @@
+#!/bin/sh
+# Runs a fast-chess (https://github.com/Disservin/fastchess) tournament based
+# on two branches of the chess_inator
+# (https://github.com/dogeystamp/chess_inator) engine.
+#
+# Example usage:
+#
+#	 cd chess_tournaments
+#	 fast-chess-branch.sh quiescence no-quiescence -openings file=8moves_v3.pgn format=pgn order=random -each tc=300+0.1 -rounds 12 -repeat -concurrency 8 -recover -sprt elo0=0 elo1=10 alpha=0.05 beta=0.05
+#
+# Do not use `main` as a branch, or any other branch already checked out in
+# another directory. You need to be in a chess_inator Git repository to run
+# this script. Ensure that the repository you're in is a throw-away worktree.
+# Create one using
+#
+# 	git worktree add ../chess_tournaments
+#
+# inside the original chess_inator repo.
+# Also, get an opening book from Stockfish's books:
+#
+# 	curl -O https://github.com/official-stockfish/books/raw/refs/heads/master/8moves_v3.pgn.zip
+#
+# The sprt mode is a statistical hypothesis testing mode that will tell you how
+# probably the first branch is better than the second branch. The Elo ratings
+# given are the "indifference zone" where the result is acceptable. To check
+# that the engine hasn't had a regression, set them to [-10, 0]. To check for
+# an improvement, use [0, 10]. Alpha and beta are probabilities for statistical
+# errors. The tournament automatically ends when a statistically significant
+# result is obtained.
+#
+# By default, a PGN file will be exported with the games played, and the
+# fast-chess SPRT output will be appended. This comment may interfere with
+# importing the PGN. But Lichess will ignore it, so it's probably fine.
+
+set -e
+
+BRANCH1="$1"
+BRANCH2="$2"
+
+# if this line fails it's because you don't have enough arguments
+shift 2
+
+COMM1=$(git rev-parse --short "$BRANCH1")
+COMM2=$(git rev-parse --short "$BRANCH2")
+
+mkdir -p games
+
+PGN=games/"$BRANCH1"__"$BRANCH2".pgn
+
+rm -f engine1 engine2
+if [ -f "$PGN" ]; then
+	rm -i "$PGN"
+fi
+
+git switch "$BRANCH1"
+cargo build --release
+cp target/release/chess_inator engine1
+
+git switch "$BRANCH2"
+cargo build --release
+cp target/release/chess_inator engine2
+
+OUTPUT=$(mktemp)
+
+fastchess \
+	-engine cmd=engine1 name="c_i $BRANCH1 ($COMM1)" \
+	-engine cmd=engine2 name="c_i $BRANCH2 ($COMM2)" \
+	-pgnout file="$PGN" \
+	timeleft=true \
+	$@ \
+	2>&1 | tee -a "$OUTPUT"
+
+printf "\n{" >> "$PGN"
+
+# match between ------- markers in fastchess output, strip newline and then output to PGN
+awk '/-{50}/{f+=1; print; next} f%2' "$OUTPUT" \
+	| head -c -1 \
+	>> "$PGN"
+
+printf "}" >> "$PGN"
+
+rm "$OUTPUT"

contrib/tl_to_clk.py

@@ -0,0 +1,39 @@
+# /usr/bin/env python
+
+"""
+Converts PGN files from fast-chess's `tl` time left notation to the standard `%clk` clock time notation.
+
+May be buggy; other comments may break this script.
+Takes stdin and converts to stdout.
+"""
+
+import sys
+import re
+
+pgn_value = sys.stdin.read()
+
+
+def convert(m: re.Match[str]) -> str:
+    seconds_total = float(m.group(1))
+    mins, secs = divmod(seconds_total, 60)
+    hrs, mins = divmod(mins, 60)
+
+    secs = round(secs, 4)
+    mins = round(mins)
+    hrs = round(hrs)
+
+    return f"{{ [%clk {hrs}:{mins:02}:{secs}] }}"
+
+pgn_value = re.sub(
+    pattern=r"{book}",
+    repl="",
+    string=pgn_value,
+)
+
+print(
+    re.sub(
+        pattern=r"{.*?tl=(.*?)s.*?}",
+        repl=convert,
+        string=pgn_value,
+    )
+)