Getting Data from Major League Baseball

During one of my grad school projects, I decided to train some AI/ML models related to baseball. In order to do this, I would need to collect any data that MLB makes available about the game scores, players, pitch by pitch results, and pitch data. While researching this, I found that since 2007, MLB has tracked every pitch thrown in games using high speed cameras and MLB makes this data available through its Gameday website as xml files available at http:/gd2.mlb.com/components/game/mlb

(Update: this portal will soon be depracated and replaced with an API. I am an updating this tool to use that API instead of the xml feed). I searched for a few tools to get the data, and by far the best tool I found available was this pitchf/x package. Unfortunately, it's written for R, and hasn't been updated in a few years. Additionally, I found that none of the tools preserved relational properties, which made it very difficult to find all pitches in sequential order thrown during a specific at bat as I would need for my AI model.

With that in mind, I decided to build my own python package. It parses the xml files into objects to simplify data interaction while preserving relational data.

Installing the package

The package is now uploaded on PyPi. Install it using pip

pip install pypitchfx

Source code is also available at https://github.com/javierpalomares90/pypitchfx

Parsing Data into Objects

The data available at MLB’s Gameday portal is formatted into xml files. There is 1 xml file per game innings_all.xml, containing all of the innings, half innings, at bats, and pitches occurring during the game formatted as xml elements in sequential order. (Take a look here for an example of the xml feed for the June 01, 2013 game between the Diamond Backs and the Cubs.) Within this file there is a game element composed of 9 inning child elements (or more if extra innings are needed to break a tie). Each inning element has a top and bottom child element for each half inning when teams swap batting and fielding. Thetop and bottom elements have atbat child elements for each batter appearance in the half inning. The child elements of the atbat element is the sequence of pitches in the at bat. Additionally, there is a players.xml file pergame containing the data for the players in the game.

The tool parses each of the xml elements into python objects for easier data manipulation.

Element_Hierarchy.png

Additionally, the tool generates unique identifiers at parse time to guarantee uniqueness and to relate parent and child elements.

gameday_er_diagram.png

Entity-Relationship diagram of the database model the pypitchfx tool writesthe Gameday data to. All of the data fields available in the xml fields are mapped tocolumns of the same name. Generated UUIDs are the tables’ primary keys. Foreign keyconstraints enforce parent child relationships. Entities maintain list of the parent to allow for fast lookups of the parent entity. The DDL statements used to insert and create thetables are available in the pypitch repository in Queries.py. Batter and Pitcher are materi-alized views to used to distinguish pitchers and batters in the Game_Player table.

Getting data objects for a given data range

Pass in a range of date (start and end inclusive) for the games you're looking for. The tool will automatically look for all games played in the given range.

from pypitchfx.scrape import scrape_games_players
games,players=scrape_games_players(start='2013-06-01',end='2013-06-01')

Writing to a database

Pass in a sqlalchemy engine and the script will write to a relational database following the ER diagram shown above.

For example:

from sqlalchemy import create_engine
from pypitchfx.scrape import scrape_games_players

engine = create_engine('postgresql+psycopg2:<user>:<pw>@localhost')
games,players=scrape_games_players(start='2013-06-01',end='2013-06-01',engine=engine)