Matrix Completion in the Unit Hypercube via Structured Matrix Factorization

This is the implementation of the approaches described in the paper:

Emanuele Bugliarello, Swayambhoo Jain, and Vineeth Rakesh. Matrix Completion in the Unit Hypercube via Structured Matrix Factorization. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19), July 2019.

We provide the code for reproducing our results, as well as pre-trained models for the Click-Through Rate (CTR) data.

This code is based on a fork of npMF. Our methods for the unit interval, Expertise Matrix Factorization (EMF) and Survival Matrix Factorization (SMF), can be found in npmf/unit_models.py.

Requirements

The requirements can be installed in either of the following ways:

• By setting up a dedicated conda environment:
  conda env create -f environment.yml followed by source activate unitmf
• With pip:
  pip install -r requirements.txt

Data Preparation

Our CTR data is extracted from the Outbrain Click Prediction competition.

Pre-processing steps and scripts to obtain the same Cross-Validation splits used in our paper are available in experiments/CTR/prepare_data/.

You can also download the final splits from Google Drive: https://drive.google.com/drive/folders/1I_NjBXwAk1aBSDdFpCh4bWxsiXTjZ5bi

Training

Scripts for training each model are provided in experiments/CTR/training_scripts/.

• Default data directory: experiments/CTR/data/
• Default checkpoints directory: experiments/CTR/checkpoints/

Best Models

The hyper-parameters used to obtain the results reported in the paper are listed in experiments/CTR/training_scripts/README.md.

You can also download the final checkpoints from Google Drive: https://drive.google.com/drive/folders/1mJWuj08ae8uBF07jb52zGu55rbT4fr-O

Evaluation

You can use the test() method in utils/eval.py to evaluate a model.

An example, along with the results reported in our paper, can be found in experiments/CTR/notebooks/results.ipynb.

License

This work is licensed under CC BY-NC 4.0. See LICENSE for details. Third-party datasets are subject to their respective licenses.
If you find our code/models or ideas useful in your research, please consider citing the paper:

@inproceedings{Bugliarello+:IJCAI2019,
  title     = {Matrix Completion in the Unit Hypercube via Structured Matrix Factorization},
  author    = {Bugliarello, Emanuele and Jain, Swayambhoo and Rakesh, Vineeth},
  booktitle = {Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, {IJCAI-19}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  pages     = {2038--2044},
  year      = {2019},
  month     = {7},
  location  = {Macao, China},
  doi       = {10.24963/ijcai.2019/282},
  url       = {https://doi.org/10.24963/ijcai.2019/282},
}