Skip to content

Load

polaris.load_dataset

load_dataset(path: str, verify_checksum: bool = True) -> Dataset

Loads a Polaris dataset.

In Polaris, a dataset is a tabular data structure that stores data-points in a row-wise manner. A dataset can have multiple modalities or targets, can be sparse and can be part of one or multiple benchmarks.

The Polaris dataset can be loaded from the Hub or from a local or remote directory.

  • Hub (recommended): When loading the dataset from the Hub, you can simply provide the owner/name slug. This can be easily copied from the relevant dataset page on the Hub.
  • Directory: When loading the dataset from a directory, you should provide the path as returned by Dataset.to_json. The path can be local or remote.

polaris.load_benchmark

load_benchmark(path: str, verify_checksum: bool = True)

Loads a Polaris benchmark.

In Polaris, a benchmark wraps a dataset with additional meta-data to specify the evaluation logic.

The Polaris benchmark can be loaded from the Hub or from a local or remote directory.

Dataset is automatically loaded

The dataset underlying the benchmark is automatically loaded when loading the benchmark.

  • Hub (recommended): When loading the benchmark from the Hub, you can simply provide the owner/name slug. This can be easily copied from the relevant benchmark page on the Hub.
  • Directory: When loading the benchmark from a directory, you should provide the path as returned by BenchmarkSpecification.to_json. The path can be local or remote.