In this repository, a classical matrix factorization algorithm has been implemented using numpy and optimizing the Mean Squared Error using Stochastic Gradient Descent. As it is a linear method, an additional version using Deep Learning techniques has been implemented in order to improve the system performance, concluding that the MSE can be decreased by an 8.5% using a very simple neural network.
Both algorithms results over a sample dataset are shown in the jupyter notebook located in ./notebooks/example.ipynb
.
Experiments can be reproduced by following the notebook. For running the code the following dependencies are needed
- Pandas
- Numpy
- Tensorflow
- Python 3.6
Matrix Factorization results
Deep Factorization results
This repository is licensed with MIT agreement. Copyright (c) 2018 Iván Vallés Pérez
For more information, please, check the LICENSE file.