Skip to content

Dataset · autogluon

Dataset for forecasting time series evaluation

autogluon/fev_datasets

Collection of real-world time series from AutoGluon for testing forecasting models.

What's inside

The repository from the AutoGluon team contains between 100 thousand and 1 million time series records (tabular, text, timeseries) for evaluating univariate and multivariate forecasting models. The main emphasis is on realistic scenarios including covariates, missing values, and other practical complexities. All datasets are brought to a single format compatible with the fev package.

What you can do with it

  • Benchmarking time series forecasting models under missing data conditions
  • Training and validation of multivariate forecasting algorithms taking into account covariates
  • Testing AutoML pipelines on realistic tabular-temporal tasks
Tasks
Time series forecasting
Size
100K–1M rows
Downloads this month
101.3K
Likes
24

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("autogluon/fev_datasets")