Skip to content

Latest commit

 

History

5 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 

Repository files navigation

Recsys Notebooks

A compilation of recommender system benchmarks.


Implementation Details

The primary plan is to utilize multiple CPU cores as much as possible and disregard the use of GPUs. Data preprocessing is done mostly with Pyspark when handling extremely large datasets. To minimize computation time, Numpy is preferred over PyTorch or Tensorflow, and is JIT compiled with Numba whenever possible.

List of implementations

1. Matrix Factorization

Reproduction of the paper Matrix Factorization for Recommender Systems. I implement from the paper equation (2), which is the pure matrix factorization with regularization(lambda=0.1), without biases or external information. I use momentum(p=0.9) to speed up stochastic gradient descent(lr=0.01). The algorithm seems to converge after 1 epoch of training(210 seconds).

Original Paper Reproduced
RMSE(k=40) 0.91 1.08

About

Implementation examples of milestone recommender models.

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages