Python scripts for data-driven paddle tennis captains and club pros
- tsrate: calculate TrueSkill ratings for players based on APTA league results
- detail2export: convert an APTA "detail" report into an "export" report suitable for use with tsrate
Paddle tennis is an increasingly popular sport, with
APTA sponsored leagues established in many
metropolitan areas in the US. The APTA has developed a player rating system
similar to the
NTRP
used in tennis, but this only goes so far in helping to answer the sorts of
questions that team captains, club pros, and club paddle chairs often find
themselves facing:
- who should play on court 1, court 2, etc?
- who should move up (or down) a series?
- in social play, how do you pair players to create an even playing field?
- how do you properly seed teams in a tournament?
There are other player assesment tools (notably the Gillespie scale), but ultimately you end up relying on the "eye test" or some other subjective measure of a player's ability. The growing ranks of players makes it unlikely that the evaluator either knows what to look for, or has watched the player enough to make an accurate assesment.
Some evaluators have started augmenting the subjective methods with some rudimentary analysis of match results to help with these decisions. The challenge that you run into almost immediately is how to deal with the wide range of player ability that exists even within a single series. How much tougher is it to win a match playing court 1, than it is playing court 2, 3, or 4? The easy solution to this problem is to try and price the courts. Creating an arbitrary weighting typically leads to self-fulfilling prophecies as the court 4 player's wins are discounted, making them appear unready to play up. Deriving a weighting scheme from the match data itself is a challenge because of variance from club to club and year to year.
The tsrate script takes a different approach, leveraging Microsoft's TrueSkill algorithm. Developed by Microsoft Research, the algorithm is used by Xbox Live games to try and match players with others at the same skill level. TrueSkill uses Bayesian inference to create a Gaussian belief distribution for each player based on the outcome of games. Notably, the algorithm supports both one-vs-one and team-based games. Heungsub Lee created an excellent implementation of TrueSkill in python (trueskill.org)
In tsrate, each set is treated as a game, and sets that go to tiebreak are counted as draws. At some level, it doesn't make a huge difference whether to rate players based on games, sets, or matches as there is a mathematical relationship between them (see: Tennis Statistics by Robert Farrenkopf) All three approaches were tested and while the outcomes were similar, using sets as the unit of scoring yields what feels like a reasonable sigma for the rating distribution based on the number of matches in a season.
What you do with the calculated ratings is ultimately up to you, but I would recommend treating it as another data point and pairing it with other methods of evaluation.
Download ZIP of this project, and unpack into a directory. From a terminal window, cd to that folder and run:
pip install -r requirements.txt
NOTE: The scripts require Python 2.6 or later. Windows users will need to install the interpreter first, if they haven't already done so. Mac and Linux users should be able to use the version that was pre-installed with their operating system.
To rate the players within an APTA paddle series, you first need a match export file. These can be generated on the myapta.org website using the reporting functionality. Filter the report to a single series, clear the club filter and apply. Download the report using the CSV link at the bottom of the page. This should give you a file called 'matches-export.csv'.
The simplest usage of tsrate is:
python tsrate.py matches-export.csv
This will generate a 'ratings.csv' file with the TrueSkill rating and sigma for every player in the series.
To display the results for a single club in the console:
python tsrate.py matches-export.csv --display <club name>
Last, but not least, to import initial ratings for some or all of the players:
python tsrate.py matches-export.csv --init <player ratings file>
Since match exports for prior years are impossible to obtain from the APTA site, you may be interested in using the detail report that's found in the 'documents' tab of the page for your series. Download the xslx file and save as a csv.
python detail2export.py <csv version of detail report> --series <series>