RecLibWH
Link to github project page: here
RecLibWH started off as a small side-project for making a custom recommender algorithm for my movie nights. Searching for the right movie is exhausting business!
It has since turned into a project where I learn about various machine learning tecniques through the lens of movie recommenders. The purpose of RecLibWH is to have a library to make available all these different machine learning algorithms.
This project has seen me going from simple nearest-neighbour collaborative filtering techniques, to sophisticated probabilistic matrix factorization techniques. Next on the to-do list is the implementation of Bayesian techniques based on a variant of latent dirichlet allocation (LDA), a popular but powerful technique used in NLP.
I also made several apps based on this library, one of the cooler-looking ones are ego-graphs centered on a single user-chosen movie, with edges drawn between itself and closely related titles, as well as edges between those titles. The following images show ego graphs for “The Big Lebowski” and “The Godfather”. (Click on them to get a better view)
For more information on how these are generated, here is the notebook I used to generate them.
Another fun project that leverages this library is FiML, this is a Django+React based browser app that I developed so I can have a fancy interface to my recommender models.