This repository contains code for the paper: LogGENE: A smooth alternative to check loss for Deep Healthcare Inference Tasks (pre-print can be found: HERE)
The /utils/ folder contains the code for the actual loss functions, with torch wrappers which allows for them to be invoked in your own work simply by doing: cirterion=loss() (where loss is TiltedLossLC if you want to use the tilted log cosh in a regression problem, sBQRq if you want to learn multiple quantiles in a binary classification problem and sBQRl, when you just want to learn a single quantile)
The Jupyter notebooks contain the code for the experiments (along with embedded results for the latest runs). The GEOLoaderFinal.ipynb notebook contains tests with the D-GEX dataset (which hasn’t been provided with the repository because of it’s size), classification.ipynb contains the code for testing the sBQC loss in datasets contained in /Datasets/Classification/, the regression.ipynb contains the code for testing the /Datasets/Regression. /D-GEX_checkpoint/ contains the losses and the model parameters of the trained network over the D-GEX dataset, the model_params can be easily loaded for further individual testing simply by using mode.load_state_dict(torch.load(<PATH TO .pt FILE>))
Please feel free to open up an issue or make a pull request in case you find any inconsistencies or want to contribute!
BIBTeX citation:
@misc{https://doi.org/10.48550/arxiv.2206.09333, doi = {10.48550/ARXIV.2206.09333},
url = {https://arxiv.org/abs/2206.09333},
author = {Jeendgar, Aryaman and Pola, Aditya and Dhavala, Soma S and Saha, Snehanshu},
keywords = {Machine Learning (cs.LG), Neural and Evolutionary Computing (cs.NE), Machine Learning (stat.ML), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {LogGENE: A smooth alternative to check loss for Deep Healthcare Inference Tasks},
publisher = {arXiv},
year = {2022},
copyright = {Creative Commons Attribution 4.0 International} }