discrete-matrix-factorization

This repository contains the source code for the paper entitled "Learning Discrete Matrix Factorization Models"

Prerequisites

- Tensorflow 
- Numpy 

The code has been tested in Ubuntu 14.04 and MacOSX, with

- Tensorflow v.1.2.1
- Numpy v.1.12.1 

You should be able to run the code with newer versions of Tensorflow and Numpy as well.

Usage

Prepare data

A sample split (75% training, 25% testing) of the MovieLens1M dataset is included (inside folder ./data/MovieLens1M/):

• rating.npz: original matrix
• train_mask.npz: training mask, same shape as the original matrix
• val_mask.npz: validation mask, same shape as the original matrix
• test_mask.npz: testing mask, same shape as the original matrix

All these matrices are saved as Numpy sparse csr_matrix.

Alternatively, you can write your own DataLoader to load and feed the model during training, with data prepared in another way.

Configurations

The configs/configs_dmfd.py file sets all the default configurations and hyper-parameters. There are some other hyperparameters inside train.py and test.py To use your own configurations, you can either edit these files, or override the default configurations using flags.

For example:

python train.py --n_epochs=1000 --initial_lr=0.01

Training

After preparing your dataset and set all the necessary hyper-parameters, you are ready to train your model. Run train.py, together with our hyper-parameters, to start the training. For example:

python train.py --data_dir=./data/MovieLens1M/ --output_basedir=./outputs/

The trained model will be saved into outputs/snapshots. To save the training log, which can be inspected using tensorboard, you need to set:

    write_summary = True

in the configuration file.

Testing

After training the model, you can test it by calling test.py, with the default parameters or with your own parameters. For example:

python test.py --data_dir=./data/MovieLens1M/ --snapshot_dir=./outputs/snapshots/

Related

If you just want to produce continuous output, you may want to check our paper "Extendable Neural Matrix Completion" (published at ICASSP 2018). The source code is available at: https://github.com/nmduc/neural-matrix-completion

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Reference

If you find the source code useful, please cite us:

D. M. Nguyen, E. Tsiligianni and N. Deligiannis, "Learning Discrete Matrix Factorization Models," 
in IEEE Signal Processing Letters, vol. 25, no. 5, pp. 720-724, May 2018.