rllm.datasets.RelF1Dataset

class rllm.datasets.RelF1Dataset(cached_dir: str, force_reload: bool | None = False)

Bases: RelBenchDataset

A wrapper for rel-f1 dataset in RelBench benchmark from RelBench: A Benchmark for Deep Learning on Relational Databases paper, which contains Formula 1 racing data with 9 tables and 3 tasks.

Tables:

• circuits

• constructor_results

• constructors

• constructor_standings

• drivers

• qualifying

• races

• results

• standings

Tasks:

• driver-dnf: Binary classification task to

  predict whether a driver did not finish a race.

• driver-position: Regression task to

  predict the finishing position of a driver.

• driver-top3: Binary classification task to

  predict whether a driver finished in the top 3.

process()

process data and save to ‘./cached_dir/{dataset}/processed/’.

property text_embedder_config: TextEmbedderConfig

Lazily create the text embedder config the first time it is needed.

Previously this was constructed as a class attribute at import time, which caused from rllm.datasets.relbench.f1 import RelF1Dataset to download/load the sentence-transformers model and took a long time.