#!/bin/bash
_S="${BASH_SOURCE[0]:-$0}"; ROOT="${ROOT:-$(cd "$(dirname "$_S")" && pwd)}"; ROOT="${ROOT%%/results/*}"; ROOT="${ROOT%%/Benchmark_eval/*}"; ROOT="${ROOT%%/Benchmarks/*}"  # RELOC_ROOT_MARK
# Rebuild the ChatTS eval requests (inference-ready) from the official Zenodo dataset_a/b.
# Adapters are fully deterministic (no seed/shuffle) — re-running gives identical output.
# dataset_a = 159 (univariate) · dataset_b = 400 (multivariate). Zenodo 10.5281/zenodo.14349206.
set -euo pipefail
PY=${PY:-/root/miniconda3/bin/python}
DIR=$ROOT/Benchmark_eval/ChatTS_test
B=$DIR/_build
DA=${DA:-$ROOT/Benchmarks/ChatTS_test/14349206/dataset_a.json}
DB=${DB:-$ROOT/Benchmarks/ChatTS_test/14349206/dataset_b.json}
[ -f "$DA" ] && [ -f "$DB" ] || { echo "raw dataset_a/b missing under $(dirname "$DA")"; exit 1; }
A() { PYTHONPATH="$B" "$PY" "$B/adapters/$1" "${@:2}" 2>&1 | tail -1; }

rm -f "$DIR"/*.jsonl

echo "== dataset_a (159, univariate) =="
A chatts_adapter.py               --input-file   "$DA" --dataset-tag dataset_a --output-file "$DIR/chatts_a_qa.jsonl"          # QA -> scored by RAGAS bridge
A chatts_align_adapter.py         --dataset-file "$DA"                          --output-file "$DIR/chatts_a_align.jsonl"       # numeric, gt = ChatTS generator attrs
A chatts_tsalign_metrics_adapter.py --dataset-file "$DA"                        --output-file "$DIR/chatts_a_tsametrics.jsonl"  # numeric, gt = TS-Align own defs (fairest)

echo "== dataset_b (400, multivariate) =="
A chatts_adapter.py               --input-file   "$DB" --dataset-tag dataset_b --output-file "$DIR/chatts_b_qa.jsonl"
A chatts_mv_adapter.py            --dataset-file "$DB"                          --output-file "$DIR/chatts_b_mv.jsonl"          # mv text (rougeL)
A chatts_mv_metrics_adapter.py    --dataset-file "$DB"                          --output-file "$DIR/chatts_b_mvmetrics.jsonl"   # mv numeric

echo "=== counts ==="
for f in "$DIR"/*.jsonl; do printf "  %-28s %s\n" "$(basename "$f")" "$(wc -l < "$f")"; done
echo "DONE -> $DIR"