Deep Reinforcement Learning in Ice Hockey for Context-Aware Player Evaluation

Authors: Guiliang Liu, Oliver Schulte

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical Evaluation shows GIM is consistent throughout a play season, and correlates highly with standard success measures and future salary.
Researcher Affiliation Academia Guiliang Liu and Oliver Schulte Simon Fraser University, Burnaby, Canada gla68@sfu.ca, oschulte@cs.sfu.ca
Pseudocode No No pseudocode or algorithm blocks are provided in the paper.
Open Source Code Yes We apply Tensorflow to implement training; our source code is published on-line.2 https://github.com/Guiliang/DRL-ice-hockey
Open Datasets No We utilize a dataset constructed by SPORTLOGi Q using computer vision techniques. The data provide information about game events and player actions for the entire 20152016 NHL (largest professional ice hockey league) season, which contains 3,382,129 events, covering 30 teams, 1140 games and 2,233 players. The paper describes the dataset and its origin but does not provide a direct link, DOI, specific citation (with authors and year), or repository information for public access.
Dataset Splits No The paper does not explicitly state training/validation/test splits (e.g., percentages, sample counts) for reproducibility.
Hardware Specification No This work was supported by an Engage Grant from the National Sciences and Engineering Council of Canada, and a GPU donation from NVIDIA Corporation. The paper mentions a 'GPU donation from NVIDIA Corporation' but does not specify the exact GPU model (e.g., RTX 2080 Ti, Tesla V100), CPU, or other hardware details used for experiments.
Software Dependencies No We apply Tensorflow to implement training; our source code is published on-line. The paper mentions 'TensorFlow' but does not specify a version number for it or any other software dependencies.
Experiment Setup Yes Our design is a 5-layer network with 3 hidden layers. Each hidden layer contains 1000 nodes, which utilize a relu activation function. The first hidden layer is the LSTM layer, the remaining layers are fully connected. We used batch size 32 (determined experimentally). Weights θ are optimized by minibatch gradient descent via backpropagation. We set tlt to the number of time steps from current time t to the beginning of the current play (with a maximum of 10 steps).