reTraining/algoritma-batch/migrate_cutoff_0600.py

"""Re-file the Jetson's stored batches under the 06:00 counting day.

`predict.py` used to turn the counting day over at 20:00, which filed the whole
day shift under the previous date — a batch that started at 08:27 on the 13th was
stored as the 12th. The archive's cycles run 06:00 to 06:00, so the two disagreed.
The default is now 06:00; this brings rows written before that change into line.

For every batch it recomputes `counting_date` from `start_time`, renumbers
`batch_number` 1..N within each counting day (per camera and object label,
ordered by start time), and rebuilds `daily_summaries` from the result.

Run it on the Jetson, against its own database:

    python migrate_cutoff_0600.py --db /opt/jetson-counter/jetson_counter.db --dry-run
    python migrate_cutoff_0600.py --db /opt/jetson-counter/jetson_counter.db

Nothing is written without a backup, and `--dry-run` writes nothing at all.
Stop `predict.py` first: it holds an active batch in memory and would write it
back under the old numbering.
"""

import argparse
import datetime
import os
import shutil
import sqlite3
import sys


def counting_date(start_time: str, cutoff_hour: int) -> str:
    """The counting day a batch belongs to, from when it started."""
    stamp = datetime.datetime.fromisoformat(start_time)
    day = stamp.date()
    if stamp.hour < cutoff_hour:
        day = day - datetime.timedelta(days=1)
    return day.isoformat()


def plan(connection, cutoff_hour: int):
    """What each row should become. Ordered by start time inside each day."""
    rows = connection.execute(
        """SELECT id, counting_date, batch_number, camera_name, object_label,
                  count, start_time
           FROM batches ORDER BY start_time"""
    ).fetchall()

    counters: dict = {}
    changes = []
    for row in rows:
        try:
            new_date = counting_date(row["start_time"], cutoff_hour)
        except (TypeError, ValueError):
            # A row whose start_time cannot be parsed is left exactly as it is;
            # guessing its day would be worse than leaving it visibly odd.
            changes.append({"row": row, "new_date": row["counting_date"],
                            "new_number": row["batch_number"], "skipped": True})
            continue
        key = (new_date, row["camera_name"], row["object_label"])
        counters[key] = counters.get(key, 0) + 1
        changes.append({"row": row, "new_date": new_date,
                        "new_number": counters[key], "skipped": False})
    return changes


def apply(connection, changes) -> None:
    """Rewrite the table.

    `batches` has UNIQUE(counting_date, batch_number, camera_name, object_label),
    so renumbering in place collides with rows that have not moved yet. The
    numbers are parked in a negative range first, which cannot collide with any
    real batch number, and then written to their final values.
    """
    cursor = connection.cursor()
    for offset, change in enumerate(changes, start=1):
        cursor.execute("UPDATE batches SET batch_number = ? WHERE id = ?",
                       (-offset, change["row"]["id"]))
    for change in changes:
        cursor.execute(
            "UPDATE batches SET counting_date = ?, batch_number = ? WHERE id = ?",
            (change["new_date"], change["new_number"], change["row"]["id"]),
        )

    cursor.execute("DELETE FROM daily_summaries")
    cursor.execute(
        """INSERT INTO daily_summaries
               (counting_date, camera_name, object_label, total_count, total_batches)
           SELECT counting_date, camera_name, object_label, SUM(count), COUNT(id)
           FROM batches
           GROUP BY counting_date, camera_name, object_label"""
    )


def main() -> int:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--db", required=True, help="path to jetson_counter.db")
    parser.add_argument("--cutoff-hour", type=int, default=6)
    parser.add_argument("--dry-run", action="store_true")
    args = parser.parse_args()

    if not os.path.isfile(args.db):
        print(f"No database at {args.db}")
        return 1

    connection = sqlite3.connect(args.db)
    connection.row_factory = sqlite3.Row
    changes = plan(connection, args.cutoff_hour)
    if not changes:
        print("No batches stored — nothing to do.")
        return 0

    moved = [c for c in changes
             if c["new_date"] != c["row"]["counting_date"]
             or c["new_number"] != c["row"]["batch_number"]]
    skipped = [c for c in changes if c["skipped"]]

    print(f"{len(changes)} batch(es) stored, {len(moved)} would change, "
          f"{len(skipped)} unparseable and left alone.\n")
    for change in moved[:20]:
        row = change["row"]
        print(f"  {row['start_time'][:19]}  "
              f"{row['counting_date']} #{row['batch_number']:<4} -> "
              f"{change['new_date']} #{change['new_number']}")
    if len(moved) > 20:
        print(f"  … and {len(moved) - 20} more")

    if args.dry_run:
        print("\nDry run — nothing written.")
        return 0
    if not moved:
        print("\nAlready consistent with the 06:00 cutoff.")
        return 0

    backup = f"{args.db}.before-0600-{datetime.datetime.now():%Y%m%d-%H%M%S}"
    shutil.copyfile(args.db, backup)
    print(f"\nBackup written to {backup}")

    try:
        with connection:
            apply(connection, changes)
    except Exception as exc:
        print(f"FAILED, database left untouched by the transaction: {exc}")
        print(f"The backup at {backup} is still the pre-migration state.")
        return 1

    days = connection.execute(
        "SELECT COUNT(DISTINCT counting_date) FROM batches").fetchone()[0]
    print(f"Done. {len(moved)} batch(es) re-filed across {days} counting day(s).")
    return 0


if __name__ == "__main__":
    sys.exit(main())