experiment_runner.download_datasets

Pre-download UCR/UEA dataset archives into the experiment cache.

When a config uses the UCR or UEA wildcard, ensure_datasets_cached is called before the experiment loop. It:

  1. Checks which datasets are already in the cache (fast no-op if all present).

  2. Downloads the full bulk archive zip in one shot when the wildcard is active.

  3. Extracts and flattens the zip so each dataset lands at {cache_dir}/{name}/{name}_TRAIN.ts — the path sktime expects via its extract_path parameter.

  4. Falls back to individual per-dataset zips for any remaining missing datasets (or when only specific datasets, not the full wildcard, are requested).

Download errors are logged as warnings and do not abort the experiment; failures will surface as FileNotFoundError or network errors when sktime tries to load the missing dataset during the experiment.

Attributes

logger

UCR_BULK_URL

UEA_BULK_URL

Functions

ensure_datasets_cached(→ None)

Ensure every dataset in dataset_names has its .ts files cached.

Module Contents

experiment_runner.download_datasets.logger
experiment_runner.download_datasets.UCR_BULK_URL = 'http://www.timeseriesclassification.com/aeon-toolkit/Archives/Univariate2018_ts.zip'
experiment_runner.download_datasets.UEA_BULK_URL = 'http://www.timeseriesclassification.com/aeon-toolkit/Archives/Multivariate2018_ts.zip'
experiment_runner.download_datasets.ensure_datasets_cached(cache_dir: pathlib.Path, dataset_names: list[str], uses_ucr_wildcard: bool = False, uses_uea_wildcard: bool = False) → None

Ensure every dataset in dataset_names has its .ts files cached.

Parameters:

  • cache_dir – Root directory where {name}/{name}_TRAIN.ts files are stored.

  • dataset_names – Fully-expanded list of dataset names (after UCR/UEA wildcard resolution).

  • uses_ucr_wildcard – True when the original config used dataset: "UCR". Triggers bulk archive download instead of individual per-dataset downloads.

  • uses_uea_wildcard – Same for dataset: "UEA".