SprinRank is a physically-inspired ranking algorithm to infer hierarchical rankings of nodes in a directed network. More details are in the paper here. Here I use SpringRank to find layer interdependence in a multi-layered network There are three jupyter notebooks provided where we test SpringRank on three different types of directed networks :
- Synthetic Network
- Tennis (2008 - 2018)
- Professional Hockey Games (NHL)
Plots are provided at the bottom in each notebook to understand SpringRank's performance against these datasets. Below is one example.
- Tennis (2008 - 2018)
The plot above shows the effect of using SpringRank to predict tennis player performances in future years on both hard and clay type surfaces. In particular, we train a model for the year 2008 and use it to predict match outcomes for the years 2009 through 2018. There two things we observe here :
- The ability to predict performance of players further down the years, gradually decreases and essentially behaves like a random model after a few years. This is expected as every year new players join and old players retire.
- SpringRank does better than randomly guessing match outcomes (red vs black dotted lines)