sreekar-voleti / Tracking-PitchControl

Working with Metrica's football tracking data and generating Pitch Control surfaces

Tracking-PitchControl

This code uses Metrica tracking data to generate Pitch Control surfaces (also called Potential Pitch Control Fields, PPCFs), as defined in the work "Beyond Expected Goals" [Spearman, 2018].

This repository assumes that the user has the Metrica tracking and event data (found here) in the directory (the default location is "../sample-data/data/").

Due to the lack of more data to train a model on, the code for the default model parameters was taken directly from Laurie Shaw's tracking data tutorial on the Friends of Tracking Data (FoTD) repo (which were taken from the original work by Spearman). The functions in Visualization.py for pitch plotting have been taken from Laurie's tutorial.

Some things to note to contrast this to other implementations of this online (here and here):

1. The use of the maximum acceleration of the players, and using a nuanced equation of motion to compute the time-to-intercept, which includes the time for the player to get to their maximum velocity.

2. The ability to have vmax and amax be player-specific.

The current state of the code allows the user to input a Game number and Frame in Main.py. For example, in Sample_Game_1 at frame 832, the code generates a figure that looks like this:

Requirements

The packages used in the code can be installed using the code below:

pip3 install numpy
pip3 install matplotlib
pip3 install pandas
pip3 install itertools 

Options (in Main.py)

1. path_to_data: Place where the Metrica data files are stored. Default is ../sample-data/data/.

2. match_name: Which game from the Metrica dataset is used for plotting. Valid inputs here are Sample_Game_X, where X is 1, 2, or 3.

3. frame: Which frame the user wants to look at.

4. field_dimen: Dimension of the field in yards. Default is (106,68).

5. include_player_velocities: Set to True to show the player velocities in the plot.

6. n_x_grid_cells: How many points along the length of the pitch to have in the grid. The corresponding number in the width dimension is determined by the ratio of dimensions in field_dimen.

7. show: Set to True to show the plot. In all cases, the plot will be saved in the Figures/ directory.

Functionality that is coming soon

1. Paralellization of PPCF evaluation on the full pitch (the fastest way is to use numba, but this package does not play well with Pandas structures and dictionaries, so there is a small amount of type-changing that needs to be done. The usual multiproecessing methods for this problem do not show a marked improvement in terms of time.)

2. Using event frame data to generate a "movie" of the events and pitch control surface evolution leading up to the event (eg. looking at the passes leading up to a goal, or analyzing buildup play).