So I’ve decided to take the plunge and have begun making repositories for my projects (old and new). I can’t believe I used to code without version control, it’s insanity! During grad school I had a neat project from my course in sparse matrix computation on Google Pagerank that I’m sharing on Github now. In my quest for making cool project names, I’ve dubbed this project LRPR(Low Rank Page Rank) and it can be found here: https://github.com/bee-rock/LRPR. Though it contains only one type of low rank approximation at the moment, perhaps that will change in the future.
The project report can be found here: The inner-outer solver for Pagerank combined with a subsampling scheme
The project starts out with a short introduction to Pagerank, formulated as en eigenvalue problem:
We investigate what the effect of a low rank approximation for the transition matrix has on the power method and an inner-outer iteration for solving the Pagerank problem. The purpose of the low rank approximation is two fold: (1) to reduce memory requirements (2) to decrease computational time. We show that we see an improvement in storage requirements and a decrease in computational time if we discard the time it takes to perform the low rank approximation, however at the sacrifice of accuracy.