Deep Weighted Averaging Classifiers

This code is to accompany the paper Deep Weighted Averaging Classifiers, by Dallas Card, Michael Zhang, and Noah A. Smith, to appear at FAT* 2019.

The repo provides support to run the DWAC and softmax models discussed in the paper, The four relevant directories for this are cifar, mnist, tabular, and text, all of which provide support for multiple datasets.

To run any of these, from the main directory, use, for example:

python -m text.run --model [basline|dwac] --dataset [dataset] --device [GPU number]

Most of the required datasets will be downloaded and preprocessed automatically.

Please use -h to see all available options.

Requirements

• python3
• pytorch 0.4
• torchvision
• numpy
• scipy
• pandas
• spacy
• scikit-learn

References

If you find this code or paper useful, please include a citation to:

• Dallas Card, Michael Zhang, and Noah A. Smith. Deep Weighted Averaging Classifiers. In Proceedings of FAT* (2019). [arXiv] [BibTex]