VIFLE

Implementation for Variational Inference using Future Likelihood Estimates ##Requirements To install requirements:

conda env create -f environment.yml
conda activate seq_vi

To run the code, download pianoroll dataset into datasets/pianoroll/(dataset) (ex. datasets/pianoroll/jsb.pkl)

Polyphonic Music Datasets

To run vifle on jsb datasets, run this command:

python seq_vi.py --pid=1

or, you can change algorithm and dataset using following command:

python seq_vi.py --algorithm=(algorithm_name) --dataset_name=(dataset_name)

(ex. python seq_vi.py --algorithm="vifle" --dataset_name="jsb")

References

If this repository helps you in your academic research, you are encouraged to cite our paper. Here is an example bibtex:

@inproceedings{KimEtal.ICML20,
  author    = {Geon-Hyeong Kim and Youngsoo Jang and Hongseok Yang and Kee-Eung Kim},
  title     = {Variational Inference for Sequential Data with Future Likelihood Estimates},
  booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
  year      = {2020}
}