UCI datasets

This repository is a lightweight fork of gpapamak/maf meant to download and process the UCI datasets of the MAF paper.

G. Papamakarios, T. Pavlakou and I. Murray, Masked Autoregressive Flow for Density Estimation, NeurIPS 2017
https://arxiv.org/abs/1705.07057

Getting started

1. Clone the uci-datasets repository.
2. Download the datasets from https://zenodo.org/record/1161203.
3. Unpack the archive.
4. Process and save the datasets as .npy files. This step requires h5py and pandas<2.0 to be installed.
5. Enjoy!

user@device:~ $ git clone https://github.com/francois-rozet/uci-datasets
user@device:~ $ cd uci-datasets
user@device:~/uci-datasets $ wget https://zenodo.org/record/1161203/files/data.tar.gz
user@device:~/uci-datasets $ tar -xzf data.tar.gz
user@device:~/uci-datasets $ python process.py

Datasets

All datasets are processed versions of public datasets.

Dataset	URL
POWER	https://archive.ics.uci.edu/dataset/235/individual+household+electric+power+consumption
GAS	https://archive.ics.uci.edu/ml/datasets/Gas+sensor+array+under+dynamic+gas+mixtures
HEPMASS	https://archive.ics.uci.edu/ml/datasets/HEPMASS
MINIBOONE	https://archive.ics.uci.edu/ml/datasets/MiniBooNE+particle+identification
BSDS300	https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/bsds/