EdgingHockeyScraper

https://pypi.python.org/pypi/edginghockeyscraper

Install

pip install edginghockeyscraper

Python Hockey Data Scraper

• Free software: MIT license
• Documentation: https://edginghockeyscraper.readthedocs.io.

Features

• Python Hockey Data Scraper with following features:

  • Caching Requests to quickly fetch data
  • Parallel Processing to speed up data fetch from NHL API

• Get League schedule for year - Usageschedule = edginghockeyscraper.get_league_schedule(2024)

• Get Game boxscore

  • boxscore = edginghockeyscraper.get_boxscore(2024020345)

• Get Game playByPlay

  • playByPlay = edginghockeyscraper.get_play_by_play(2024020345)

• Get Season boxscores

  • boxscoreSeason = edginghockeyscraper.get_boxscore_season(2024)

• Get Season playByPlay

  • playByPlaySeason = edginghockeyscraper.get_play_by_play_season(2024)

Filter Season data by PreSeason, Regular, PostSeason gametypes

• e.g.
  • games = edginghockeyscraper.get_league_schedule(2024, {GameType.REG})
  • boxscoreSeason = edginghockeyscraper.get_boxscore_season(2024, {GameType.REG})

Fetch multiple seasons in one pooled call

• Season-range variants of the fetchers above pull the schedule for each season, then issue a single pooled fetch across every game in the range instead of spinning up a new worker pool per season:
  • boxscoreSeasons = edginghockeyscraper.get_boxscore_seasons(range(2020, 2025))
  • playByPlaySeasons = edginghockeyscraper.get_play_by_play_seasons(range(2020, 2025))
  • shiftsSeasons = edginghockeyscraper.get_shifts_seasons(range(2020, 2025))
  • onIcePbpSeasons = edginghockeyscraper.get_on_ice_players_with_play_by_play_seasons(range(2020, 2025))
  • stintsSeasons = edginghockeyscraper.build_stints_seasons(range(2020, 2025))
• Prefer these over looping the single-season functions when backfilling a range of seasons (e.g. building an xG training set) -- one pool for the whole range instead of one per season.

Choose a fetch backend

• All season and multi-season fetchers accept fetch_backend ('process' or 'thread', default 'process') and max_workers:
  • boxscoreSeason = edginghockeyscraper.get_boxscore_season(2024, fetch_backend='thread', max_workers=16)
• 'process' matches historical behavior (CPU-isolated workers), and suits cases with heavier per-game post-processing (e.g. build_stints_season/build_stints_seasons).
• 'thread' is often faster for the pure single-endpoint fetchers (get_boxscore, get_play_by_play, get_shifts, get_on_ice_players_with_play_by_play) since each call is a blocking HTTP GET + JSON parse -- I/O-bound work that releases the GIL while waiting on the network, and threads skip the cost of pickling large payloads back across a process boundary. Benchmark on your own connection/CPU before assuming thread is faster -- it depends on how much the NHL API rate-limits concurrent connections, and requests-cache's sqlite backend serializes writes from many threads in one process, which can become the bottleneck at high thread counts.
• max_workers=None (the default) keeps each backend's own default (process_map -> os.cpu_count(); thread_map -> min(32, os.cpu_count() + 4)).

Utilize requests-cache for fast repeated request calls

• Caching is on by default; pass disable_cache=True to bypass it.

• First Call:

  %%time
  edginghockeyscraper.get_boxscore_season(2024)
  

  CPU times: user 583 ms, sys: 318 ms, total: 901 ms Wall time: 1min 18s

• Second Call:

  • CPU times: user 374 ms, sys: 141 ms, total: 515 ms Wall time: 1.31 s

  A 60x speedup!

Utilize multiprocessing to improve request speed

Benchmark using 2024 Macbook Air Apple M3 16GB

1. No Parallel getBoxscoreSeason: CPU times: user 15.6 s, sys: 3.55 s, total: 19.1 s Wall time: 8min 49s

2. Parallel getBoxscoreSeason: CPU times: user 583 ms, sys: 318 ms, total: 901 ms Wall time: 1min 18s

   A ~7x Speedup! (this is an 8-core CPU - you can expect roughly a <# cpu-cores> speedup)