nodegam 0.1.0

NodeGAM - an interpretable deep learning GAM model.

NODE GAM: Differentiable Generalized Additive Model for Interpretable Deep Learning:

This repository is the official implementation of NODE GAM: Differentiable Generalized Additive Model for Interpretable Deep Learning.

Requirements

To install the package, run

pip install nodegam

And follow the notebooks or the main file in this repo.

Or you can install requirements by:

pip install -r requirements.txt

Training

Please see the following colab notebook for how to train and visualize a NODE-GA2M on Bikeshare. https://colab.research.google.com/drive/1C_gBoSc1AlQ7VvCXVWiU-7X3YjQZTiZI?usp=sharing

And see more examples under notebooks/

We provide the hyperparmeters we use in best_hparams/.
To reproduce our results, e.g. NODE-GA2M trained in fold 0 (total 5 folds) of bikeshare, you can run

python main.py --name 0603_best_bikeshare_f0 --load_from_hparams resources/best_hparams/node_ga2m/0519_f0_best_bikeshare_GAM_ga2m_s83_nl4_nt125_td1_d6_od0.0_ld0.3_cs0.5_lr0.01_lo0_la0.0_pt0_pr0_mn0_ol0_ll1 --fold 0

The models will be stored in logs/0603_best_bikeshare_f0/. And the results including test/val error are stored in results/bikeshare_GAM.csv

Baseline

You can train Spline and EBM by the following two commands.

python baselines.py --name 0603_bikeshare_spline_f0 --fold 0 --model_name spline --dataset bikeshare
python baselines.py --name 0603_bikeshare_ebm_f0 --fold 0 --model_name ebm-o100-i100 --dataset bikeshare

The result is shown in results/baselines_bikeshare.csv and the model is stored in logs/0603_bikeshare_spline_f0/.

Visualization

To visualize the model, run this in a notebook:

from nodegam.gams.vis_utils import vis_main_effects
from nodegam.utils import average_GAMs

df_dict = {
    'node_ga2m': average_GAMs([
        '0603_best_bikeshare_f0',
    ], max_n_bins=256),
    'ebm': average_GAMs([
        '0603_bikeshare_ebm_f0',
    ], max_n_bins=256),
}

fig, ax = vis_main_effects(df_dict)

Note max_n_bins above is to only calculate the GAM on at most 256 points per feature. This avoids too-long computations.

See bikeshare visualization.ipynb which we provide all graphs for all models (NODE-GA2M, NODE-GAM, EBM and Spline) in our paper, and you can see files in best_dfs/.

Results

See the Table 1 and 2 of our paper.

Citations

If you find the code useful, please cite:

@inproceedings{chang2021node,
  title={NODE-GAM: Neural Generalized Additive Model for Interpretable Deep Learning},
  author={Chang, Chun-Hao and Caruana, Rich and Goldenberg, Anna},
  booktitle={International Conference on Learning Representations},
  year={2021}
}

Contributing

All content in this repository is licensed under the MIT license.