This is TensorFlow implementation for the paper "Deep N-ary Error Correcting Output Codes" (MobiMedia 2020): https://arxiv.org/pdf/2009.10465.pdf.

• python 3.x with tensorflow (1.8.0~1.13.1), keras, numpy, sklearn, scipy, matplotlib, tqdm.

• TREC: Question-type classification dataset (6-classes), [link].
• STS-5: Stanford Sentiment Treebank dataset (5-classes), [link].

Convert file format to UTF-8 on Mac OS X: iconv -f <other_format> -t utf-8 file > new_file
Convert file format to UTF-8 on Ubuntu Linux: iconv -f <other_format> -t utf-8 file -o new_file

• MNIST: Handwritten digit dataset (10-classes), [link].
• CIFAR: Real-world image dataset (10/100-classes), [link].
• FLOWER-102: 102 Category Flower Dataset (102-classes), [link] or [download]. (FLOWER-102 utilizes the pretrained AlexNet, which can be downloaded here: [kratzert/finetune_alexnet_with_tensorflow]).

Take the MNIST dataset as an example. Training the model for MNIST dataset without sharing base learners' parameters:

python train_mnist.py --gpu_idx 0 \  # specify the gpu index
                      --training True \  # specify the mode (training or testing)
                      --num_meta_class 3 \  # specify the number of meta classes
                      --num_classifier 60  # specify the number of base learners

Training the model for MNIST dataset with full sharing base learners' encoder parameters (except for the classifier head):

python train_mnist_full.py --gpu_idx 0 \  # specify the gpu index
                           --training True \  # specify the mode (training or testing)
                           --num_meta_class 3 \  # specify the number of meta classes
                           --num_classifier 60  # specify the number of base learners

If you feel this project helpful to your research, please cite our work.

@article{zhang2020deep,
  title={Deep N-ary Error Correcting Output Codes},
  author={Zhang, Hao and Zhou, Joey Tianyi and Wang, Tianying and Tsang, Ivor W and Goh, Rick Siow Mong},
  journal={arXiv preprint arXiv:2009.10465},
  year={2020},
  url={https://arxiv.org/pdf/2009.10465.pdf}
}

and

@article{zhou2019n,
  title={N-ary decomposition for multi-class classification},
  author={Zhou, Joey Tianyi and Tsang, Ivor W and Ho, Shen-Shyang and M{\"u}ller, Klaus-Robert},
  journal={Machine Learning},
  volume={108},
  number={5},
  pages={809--830},
  year={2019},
  publisher={Springer}
}