Skip to content

Latest commit

 

History

History
37 lines (22 loc) · 1.65 KB

README.md

File metadata and controls

37 lines (22 loc) · 1.65 KB

Extensible Matrix Factorization (EMF)

GitHub license

EMF schematic diagram

Beta release EMF and other baselines are implemented with TensorFlow v1.

Setup:

We recommend users to install required packages dependencies for the EMF models and experiments, using Conda. To install Python and other dependencies, which you can do directly using the provided environment.yml file:

conda env create -f environment.yml
source activate EMF

The emf Conda environment can be used to run any number of experiments and examples included in this project/repository. We note that experiments and data download for other parts of this project are implemented and configured with Snakemake which will be installed as part of the EMF environment.

To run models using GPUs:

To use GPUs, install the emf-gpu environment via:

conda env create -f environment-gpu.yml
source activate EMF-gpu

Experiments:

10 fold Monte-Carlo cross-validation benchmarks of matrix factorization models

See experiments/all-mf-monte-carlo-cv.

Hyperparameter search with hyperopt

Hyperparameter searches for different MF models are implemented and can be run using snakemake in the following directories in experiments/:

  • single-species-mf-hp-search - MF, KPMF, MF with bias, KPMF with bias, NGMC models.
  • xsmf-hp-search - XSMF model
  • k-xsmf-hp-search - Kernelized XSMF model
  • k-xsmf-b-hp-search - Kernelized XSMF with bias model