Welcome to this repository, where you'll find a PyTorch implementation of Variable-Selection Network (VSN). VSN is a deep neural network architecture designed to learn and select relevant input variables, enhancing both model interpretability and performance. I chose PyTorch for VSN implementation due to the absence of an existing version online. The closest reference is a Keras example titled "Classification with Gated Residual and Variable Selection Networks".
The VSN architecture is inspired by the paper "Gated Residual and Variable Selection Networks for Tabular Data" and the winning solution of the Kaggle competition "ICR - Identifying Age-Related Conditions", which utilized VSNs effectively.
Experiments utilized the datasets from the Kaggle competition "ICR - Identifying Age-Related Conditions" & "Prediction of spam with Bayesian model". The VSN implementation performed well in this competition.
| Dataset | Access Link |
|---|---|
| ICR - Identifying Age-Related Conditions | ICR |
| Prediction of Spam with Bayesian Model | Spam |