File size: 3,518 Bytes
2e511b5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 | #!/usr/bin/env bash
# Reproduce every generated table and figure of the LEGEX paper from the
# published JSONL bundles (goldensets + inference-results).
#
# Inputs (auto-detected, or override via env):
# GOLD_DIR goldenset data dir (submission/goldensets/data, else ../goldensets/data)
# INF_DIR inference data dir (submission/inference-results/data, else ../inference-results/data)
# ALTTEST_DIR optional clone of https://github.com/nitaytech/AltTest — reruns
# the Alternative Annotator Test (alt_test_pooled.csv + the
# per-field reference CSVs; skipped when unset)
#
# Every artifact is regenerated in place under data/analysis/, so a clean
# `git status` after a run is the verification that the shipped numbers
# reproduce.
set -euo pipefail
cd "$(dirname "$0")/.."
if [[ -z "${GOLD_DIR:-}" ]]; then
for c in goldensets/data submission/goldensets/data ../goldensets/data; do
[[ -d "$c" ]] && GOLD_DIR="$c" && break
done
fi
if [[ -z "${INF_DIR:-}" ]]; then
for c in inference-results/data submission/inference-results/data ../inference-results/data; do
[[ -d "$c" ]] && INF_DIR="$c" && break
done
fi
if [[ -z "${GOLD_DIR:-}" || -z "${INF_DIR:-}" ]]; then
echo "error: published data not found — clone legexbenchmark/goldensets and" >&2
echo "legexbenchmark/inference-results next to this repository, or set GOLD_DIR/INF_DIR." >&2
exit 1
fi
echo "gold: $GOLD_DIR"
echo "inference: $INF_DIR"
RUN="uv run --extra plots"
IAA_DIR="data/analysis/iaa"
# Byte-stable matplotlib PDF metadata across runs.
export SOURCE_DATE_EPOCH=0
# Clean first so dropped inputs can't leave stale rows behind. The AAT CSVs
# are only regenerated when ALTTEST_DIR is set (they need the upstream clone).
rm -f "$IAA_DIR"/pairwise_agreement.csv "$IAA_DIR"/kappa_audit.csv \
"$IAA_DIR"/ANALYSIS.md "$IAA_DIR"/alt_test_decomposition.csv
echo "== Table 3 inputs: human-human agreement (pairwise_agreement.csv, kappa_audit.csv)"
$RUN legex-iaa --gold-dir "$GOLD_DIR" --out "$IAA_DIR"
echo "== Table 5: AAT win/tie/loss decomposition (alt_test_decomposition.csv)"
$RUN python scripts/alt_test_decomposition.py \
--gold-dir "$GOLD_DIR" --inference-dir "$INF_DIR" --out "$IAA_DIR"
echo "== Scoring aggregates (per_*.csv, tables/headline.tex, tables/per_field.tex)"
$RUN legex-analysis --gold-dir "$GOLD_DIR" --inference-dir "$INF_DIR" \
--country am au be br ch de es fr ge hk in np nz ph rs sg tw uk us
echo "== Table 4: headline metrics (quant_results.tex)"
$RUN legex-quant-results
echo "== Tables 7/8: by-jurisdiction and by-field metrics (paper_tables.tex)"
$RUN python scripts/paper_tables.py > data/analysis/paper_tables.tex
echo "== Table 9 + word cloud: diversity stats"
$RUN python scripts/diversity_stats.py --gold-dir "$GOLD_DIR"
echo "== Table 10 + Figure 5: currency and ISIC frequencies"
$RUN python scripts/appendix_frequencies.py --gold-dir "$GOLD_DIR"
echo "== Figure 3: hallucination category shares"
$RUN python scripts/hallucination_stats.py
if [[ -n "${ALTTEST_DIR:-}" ]]; then
echo "== AAT (alt_test_pooled.csv + alt_test_reference_*.csv)"
$RUN python scripts/alt_test_reference.py --alttest "$ALTTEST_DIR" --per-field \
--gold-dir "$GOLD_DIR" --inference-dir "$INF_DIR" --out "$IAA_DIR"
else
echo "== AAT skipped (set ALTTEST_DIR to a clone of nitaytech/AltTest to rerun it)"
fi
echo "== ANALYSIS.md (renders all CSVs above)"
$RUN legex-analysis-report
echo "done — verify with: git status data/analysis"
|