v5.0 Dataset — the Numerai Tournament

"Atlas" released in July 2024. This dataset alters and expands the dataset Universe to include more useful stocks. All files are only in int8 formats.

train.parquet

Data used to train your model

validation.parquet

Data used to validate or train your model. This expands every week. Contains targets for eras that in the V3 dataset were marked as test and didn't previously have targets provided. Expands every week to include data for the most recent eras with targets available.

validation_example_preds.parquet

An example file with predictions on the validation.parquet file. Not fully supported by our diagnostics tool yet.

live.parquet

The live data that your model predicts on. This changes daily.

live_example_preds.parquet

An example file with predictions on the live.parquet file. Allows you to easily try your first submission.

features.json

JSON file with statistics about each feature in the dataset and some pre-made feature sets to make modeling easier.

train_benchmark_models.parquet

A file with benchmark model predictions for some of the train data.

validation_benchmark_models.parquet

A file with all benchmark model predictions for validation data.

live_benchmark_models.parquet

A file with all benchmark model predictions for live data.

meta_model.parquet

The Meta Model gives you insights on how your model would have altered the Meta Model historically, as well as how closely your predictions align with or diverge from the consensus signal, helping you optimize for metrics like MMC. This version of the Meta Model is only available from era 1133 onwards.