Code and data for the paper "Coherent Hierarchical Multi-label Classification Networks".

In order to evaluate the model for a single seed run:

  python main.py --dataset <dataset_name> --seed <seed_num> --device <device_num>

Example:

  python main.py --dataset cellcycle_FUN --seed 0 --device 0

Note: the parameter passed to "dataset" must end with: '_FUN', '_GO', or '_others'.

If you want to execute the model for 10 seeds you can modify the script main_script.sh and execute it.

The results will be written in the folder results/ in the file <dataset_name>.csv.

If you want to execute again the hyperparameters search you can modify the script script.shaccording to your necessity and execute it.

The code was run on a Titan Xp with 12GB memory. A description of the environment used and its dependencies is given in c-hmcnn_enc.yml.

By running the script main_script.sh we obtain the following results (average over the 10 runs):

@inproceedings{giunchiglia2020neurips,
    title     = {Coherent Hierarchical Multi-label Classification Networks},
    author    = {Eleonora Giunchiglia and
               Thomas Lukasiewicz},
    booktitle = {34th Conference on Neural Information Processing Systems (NeurIPS 2020)},
    address = {Vancouver, Canada},
    month = {December},
    year = {2020}
}