fixing dataviewer issue

manifests/dev.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6c2d60a2423924718d12df67fdddf10ee04e8d818271fe2f5183cfa0207a089
-size 11675
+oid sha256:d71f9cae09d6310900e09728373f0b7422481f872ad117f7a2fb2cf12a9e6c00
+size 8297

manifests/preview.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93325e2fef00b641d5dae45a4b85423bf139fe964e27810d58386f268efa8eb2
-size 9765
+oid sha256:90bcb110dcc8d1b54b2ec4ab071a3059e5b235959fb3c6fafb41a2f7c047c1cd
+size 6180

scripts/jsonl_to_parquet_arrow.py

@@ -0,0 +1,90 @@
+#!/usr/bin/env python3
+"""
+Convert JSONL to Parquet using an explicit Arrow schema suitable for the MMDN dataset.
+
+Usage:
+  python scripts/jsonl_to_parquet_arrow.py manifests/preview.jsonl manifests/preview.parquet
+  python scripts/jsonl_to_parquet_arrow.py manifests/dev.jsonl manifests/dev.parquet
+"""
+import argparse
+import json
+from pathlib import Path
+from typing import List, Dict, Any
+
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+
+SCHEMA = pa.schema([
+    pa.field("parcellation", pa.string()),
+    pa.field("subject", pa.string()),
+    pa.field("corr_path", pa.string()),
+    pa.field("ts_path", pa.string()),
+    pa.field("correlation_matrix", pa.list_(pa.list_(pa.float32()))),
+    pa.field("corr_shape", pa.list_(pa.int64())),
+    pa.field("ts_shape", pa.list_(pa.int64())),
+    pa.field("corr_sha256", pa.string()),
+    pa.field("corr_bytes", pa.int64()),
+    pa.field("ts_sha256", pa.string()),
+    pa.field("ts_bytes", pa.int64()),
+])
+
+
+def load_jsonl(path: Path) -> List[Dict[str, Any]]:
+    rows: List[Dict[str, Any]] = []
+    with path.open("r", encoding="utf-8") as f:
+        for line in f:
+            if not line.strip():
+                continue
+            rows.append(json.loads(line))
+    return rows
+
+
+def to_parquet(rows: List[Dict[str, Any]], dst: Path):
+    # Ensure all fields exist in each row
+    for r in rows:
+        r.setdefault("parcellation", "")
+        r.setdefault("subject", "")
+        r.setdefault("corr_path", "")
+        r.setdefault("ts_path", "")
+        r.setdefault("correlation_matrix", [])
+        r.setdefault("corr_shape", [])
+        r.setdefault("ts_shape", [])
+        r.setdefault("corr_sha256", "")
+        r.setdefault("corr_bytes", 0)
+        r.setdefault("ts_sha256", "")
+        r.setdefault("ts_bytes", 0)
+
+    # Build columns with explicit types
+    cols = {}
+    cols["parcellation"] = pa.array([r["parcellation"] for r in rows], type=pa.string())
+    cols["subject"] = pa.array([r["subject"] for r in rows], type=pa.string())
+    cols["corr_path"] = pa.array([r["corr_path"] for r in rows], type=pa.string())
+    cols["ts_path"] = pa.array([r["ts_path"] for r in rows], type=pa.string())
+    cols["correlation_matrix"] = pa.array([r["correlation_matrix"] for r in rows], type=pa.list_(pa.list_(pa.float32())))
+    cols["corr_shape"] = pa.array([r["corr_shape"] for r in rows], type=pa.list_(pa.int64()))
+    cols["ts_shape"] = pa.array([r["ts_shape"] for r in rows], type=pa.list_(pa.int64()))
+    cols["corr_sha256"] = pa.array([r["corr_sha256"] for r in rows], type=pa.string())
+    cols["corr_bytes"] = pa.array([int(r["corr_bytes"]) if r["corr_bytes"] is not None else 0 for r in rows], type=pa.int64())
+    cols["ts_sha256"] = pa.array([r["ts_sha256"] for r in rows], type=pa.string())
+    cols["ts_bytes"] = pa.array([int(r["ts_bytes"]) if r["ts_bytes"] is not None else 0 for r in rows], type=pa.int64())
+
+    table = pa.table(cols, schema=SCHEMA)
+    dst.parent.mkdir(parents=True, exist_ok=True)
+    pq.write_table(table, dst)
+    print(f"Wrote {dst} with schema: {table.schema}")
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("src", type=Path)
+    ap.add_argument("dst", type=Path)
+    args = ap.parse_args()
+
+    rows = load_jsonl(args.src)
+    to_parquet(rows, args.dst)
+
+
+if __name__ == "__main__":
+    main()
+