Using Random Effects to Account for High-Cardinality Categorical Features and Repeated Measures in Deep Neural Networks
This is the working directory for our Neurips 2021 paper.
For full implementation details see the paper and supplemental.
For running the simulations use the simulate.py file, like so:
python simulate.py --conf conf_files/conf_random_intercepts.yaml --out res.csv
The --conf attribute accepts a yaml file such as conf_random_intercepts.yaml which you can change.
To run various real data experiments see the jupyter notebooks in the notebooks folder. We cannot unfortunately attach the actual datasets, see paper for details.
For using LMMNN with your own data use the NLL loss layer as shown in notebooks and simulation.