nba_games (TsFile)

Apache TsFile version of choucsan/NBA_Games.

Overview

Metadata, official box scores and official play-by-play annotations that link full-length NBA game videos on YouTube to NBA.com game data. Instead of redistributing video, the dataset provides the YouTube video id/url for each retained game together with the cleaned matchup, the game date, the official box score and — when NBA.com exposes it — the official play-by-play action sequence.

• Tables: 4 TsFile table(s), 86,927 rows in total.
  • box_scores: 5,194 rows, 1970-05-08 00:00:00 → 2025-02-03 00:00:00.029000 (UTC)
  • game_metadata: 189 rows, 1970-01-01 00:00:00 → 1970-01-01 00:00:00.188000 (UTC)
  • games: 189 rows, 1970-05-08 00:00:00 → 2025-02-03 00:00:00 (UTC)
  • play_by_play: 81,355 rows, 1996-11-01 00:00:00 → 2025-02-03 00:00:00.517000 (UTC)

Schema (TsFile structure)

box_scores.tsfile

• Time (INT64, milliseconds) — the source game_date column.
• TAG: none — the table is stored as a single device.
• FIELD (42): source (string), nba_game_url (string), youtube_id (string), youtube_url (string), row_type (string), game_id (string), side (string), team_id (int64), team_city (string), team_name (string), team_tricode (string), team_slug (string), score (double), periods (string), statistics_minutes (string), statistics_fieldGoalsMade (int64), statistics_fieldGoalsAttempted (int64), statistics_fieldGoalsPercentage (double), statistics_threePointersMade (int64), statistics_threePointersAttempted (int64), statistics_threePointersPercentage (double), statistics_freeThrowsMade (int64), statistics_freeThrowsAttempted (int64), statistics_freeThrowsPercentage (double), statistics_reboundsOffensive (int64), statistics_reboundsDefensive (int64), statistics_reboundsTotal (int64), statistics_assists (int64), statistics_steals (int64), statistics_blocks (int64), statistics_turnovers (int64), statistics_foulsPersonal (int64), statistics_points (int64), statistics_plusMinusPoints (int64), person_id (double), first_name (string), family_name (string), name_i (string), player_slug (string), position (string), jersey_num (string), comment (string)

game_metadata.tsfile

• Time (INT64, milliseconds) — the source game_date column.
• TAG: none — the table is stored as a single device.
• FIELD (16): source_game_id (string), source_game_url (string), source_game_title (string), source_game_date (string), source_game_duration (int64), source_game_description (string), official_game_game_id (string), official_game_game_date (string), official_game_away_abbr (string), official_game_home_abbr (string), official_game_game_url (string), official_game_box_score_url (string), official_game_play_by_play_url (string), folder (string), box_score_rows (int64), play_by_play_rows (int64)

games.tsfile

• Time (INT64, milliseconds) — the source date column.
• TAG: none — the table is stored as a single device.
• FIELD (5): id (string), url (string), title (string), duration (int64), description (string)

play_by_play.tsfile

• Time (INT64, milliseconds) — the source game_date column.
• TAG: none — the table is stored as a single device.
• FIELD (28): source (string), game_id (string), nba_game_url (string), youtube_id (string), youtube_url (string), actionNumber (int64), clock (string), period (int64), teamId (int64), teamTricode (string), personId (int64), playerName (string), playerNameI (string), xLegacy (int64), yLegacy (int64), shotDistance (int64), shotResult (string), isFieldGoal (int64), scoreHome (string), scoreAway (string), pointsTotal (int64), location (string), description (string), actionType (string), subType (string), videoAvailable (int64), shotValue (int64), actionId (int64)

Conversion notes

• The source is JSONL/JSON, so it is flattened to four tables before the TsFile write: games (one row per retained YouTube game), box_scores (one row per team/player box-score line, 189 games), play_by_play (81,355 official play-by-play actions over 166 games) and game_metadata (one row per game folder).
• Time = the game date as INT64 milliseconds at 00:00 UTC (date / game_date). Within a game the rows are ordered by a 1 ms step in stable source order; the game clock itself is kept as the period and clock FIELDs.
• Nested JSON objects are flattened: the per-team statistics block becomes stat_* columns and periods is stored as JSON text; affected_services-style null columns are absent from the source schemas here.
• No column of the flattened tables is dropped. The 23 games whose play-by-play file is empty contribute box-score rows only, exactly as in the source.
• Not converted: games/*/video/ — the repository only carries empty placeholders; no video is redistributed by the source, and none is downloaded or uploaded here.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("box_scores.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: choucsan/NBA_Games
• License: mit (as declared by the original dataset).