Reparameterizable Subset Sampling via Continuous Relaxations

This repo contains the code for the paper Reparameterizable Subset Sampling via Continuous Relaxations, which allows you to include subset sampling as a layer in a neural network. This is useful whenever you want to select a discrete number of elements, such as in dynamic feature selection or k-nearest neighbors. This repo contains the experiments for learning feature selectors for explainability, training a deep stochastic k-NN model, and training a parametric t-SNE model using subset sampling.

Supports the following libraries:

• PyTorch (SubsetOperator in subsets/knn/sorting_operator.py)
• TensorFlow (sample_subset in subsets/sample_subsets.py)

To setup, please create a new Python virtualenv with Python 3.6, activate it, navigate to this directory (containing setup.py) and run pip install -e .

To run the experiments, navigate to the experiments/ folder and run the corresponding scripts.

If you find this code useful, please cite

@article{xie2019subsets,
  author    = {Sang Michael Xie and Stefano Ermon},
  title     = {Reparameterizable Subset Sampling via Continuous Relaxations},
  journal   = {International Joint Conference on Artificial Intelligence (IJCAI)},
  year      = {2019}
}