opringle / collaborative_filtering

A repository to experiment building state of the art recommender systems using apache MXNet

Deep Learning Recommender Systems

A repository to experiment building recommender systems using apache MXNet.

Repo structure

• /data directory used for storing model input data
• /libs python modules used by model training script
• ./recysys.py trains the recommender system
• ./results directory used for storing model reports and results
• ./docs: research notes etc

Running the code

1. Clone this repo and cd to root
2. Download, unzip and store the ratings data:
   1. $ mkdir data
   2. $ wget 'http://files.grouplens.org/datasets/movielens/ml-1m.zip'
   3. $ unzip -a ml-1m.zip
   4. $ mv ./ml-1m/ratings.dat ./data/
   5. $ rm -rf ml-1m
   6. $ rm ml-1m.zip
3. Train the model: $ python recsys.py

ToDo

• Use normal data iterators with data in memory
• Use callback functions to reproduce HR@K plots