UCI datasets

Regression datasets from the UCI machine learning repository prepared for benchmarking studies with test-train splits.

Installation

Install using pip (the download size is about 312 Mb):

python -m pip install git+https://github.com/treforevans/uci_datasets.git

Usage

The following code gets the first test-train split (i.e., split=0) of the challenger dataset:

from uci_datasets import Dataset
data = Dataset("challenger")
x_train, y_train, x_test, y_test = data.get_split(split=0)

There are 10 test-train splits for each dataset (as in 10-fold cross validation) with 90% of the dataset being training points and 10% being testing points in each split. The split parameter of the Dataset.get_split method accepts integers from 0 to 9 (inclusive).

Datasets

The below table contains the size (number of observations) and the number of input dimensions of each dataset. All datasets have a single output dimension.

Dataset information can be obtained from the all_datasets dictionary. For example, to obtain a list of all datasets with fewer than 1000 observations, execute the following:

from uci_datasets import all_datasets
[name for name, (n_observations, n_dimensions) in all_datasets.items() if n_observations < 1000]

Papers using these datasets

The following papers use the same datasets and test-train splits present in this repository.

• Yang, Zichao, et al. "A la carte–learning fast kernels." Artificial Intelligence and Statistics. 2015.
• Wilson, Andrew Gordon, et al. "Deep kernel learning." Artificial Intelligence and Statistics. 2016.
• Evans, Trefor W., and Prasanth B. Nair. "Scalable Gaussian processes with grid-structured eigenfunctions (GP-GRIEF)." International Conference on Machine Learning. 2018.
• Evans, Trefor W., and Prasanth B. Nair. "Discretely relaxing continuous variables for tractable variational inference." Neural Information Processing Systems. 2018.
• Audouze, Christophe, and Prasanth B. Nair. "Sparse low-rank separated representation models for learning from data." Proceedings of the Royal Society A 475.2221 (2019).
• Evans, Trefor W., and Prasanth B. Nair. "Quadruply stochastic gaussian processes." arXiv preprint arXiv:2006.03015 (2020).