A tensorflow implementation of a human-in-the-loop feature selection (HILFS) architecture introduced in this AAAI 2019 paper.
The code available here reproduces the image classification experiments presented on that paper.
- python (Verified on 3.6.0, not tested on Python 2)
- numpy
- pandas
- sklearn
- tensorflow (version 1.1.0)
- tqdm
A complete list of the arguments can be found at the main.py file.
The model can be run as follows
python main.py -d directory_path -m type_of_model -e number_of_epochs -nc number_of_categories
All the results are automatically logged in the directory defined by the -d argument.
They can be checked by running tensorboard and opening the browser on the localhost:6006
tensorboard --logdir=directory_path
If you find HILFS useful please cite us in your work:
@inproceedings{Correia2019,
author = {Correia, Alvaro H. C. and Lecue, Freddy},
booktitle = {Thirty-Third AAAI Conference on Artificial Intelligence},
title = {Human-in-the-Loop Feature Selection},
year = {2019}
}