Skip to content

sfu-cl-lab/martin-project

Repository files navigation

martin-project

Martin's MSc project repository

Dataset

The dataset contains play-by-play data from all manually tagged UBC Men's Volleyball games in 2017/2018, 2018/2019 and 2019/2020 seasons. Additionally, a number of games between other teams in Canada West are included as they were used for game planning for UBC.

There are two versions of the dataset: one where locations are described categorically using zones (Z1, Z2, ...) and one where they are represented with X and Y coordinates.

Columns

  • Season
  • GameID
  • PlayerTeam: team executing last action
  • PlayerName: player executing last action
  • RewardValue, RewardDistance: +1 for home point, -1 for away point, RewardDistance contains distance to end of current rally
  • SetNumber: current set (1-5)
  • ScoreMax: maximum of the home and away scores in current set (eg. for 21-18 it would be 21)
  • ScoreDiff: away score minus home score in current set (eg. for 21-18 it would be -3)
  • ActionHome, ActionAway: flags for action performed by home or away team
  • ActionType: one of Serve, Receive, sEt, Attack, Block, Dig or Free-ball
  • ActionStartZone, ActionEndZone: start and end location of action performed
  • ActionSpeed: flag denoting a spin serve or a fast set (vs. float serve or high set)
  • ActionOutcome: one of the possible action outcomes: = (error), -, /, !, +, # (perfect)

To incorporate action history, up to 10 most recent actions (from beginning of rally) are included in each data point.

Notebooks:

  • DecisionTree: contains experiments with fitting a decision tree classifier to the dataset, predicting whether the point will go to the home team (1) or away team (-1). Prediction probabilities also considered using ratio of training data in leaves. 5-fold cross validation is used to prevent overfitting.
  • RandomForest: same as above but using a random forest classifier.
  • I have also added several other notebooks such as NeuralNet and MimicTree that experiment with other models.
  • Questions: contains various example analyses using values from the reinforcement learning neural net.

RL Code

  • Them nn_code subfolder contains Python code for the reinforcement learning neural network.

About

Martin's MSc project

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published