cuperiod.batch_periodograms

cuperiod.batch_periodograms(inputs, method='GLS', *, backend='auto', device='cpu', workers=None, grid=None, settings=None, columns=None, domain=None, band_column=None, sink=None, n_best=10, store_raw=False, resume=True, chunk_size=256)[source]

Compute periodograms for many light curves.

Parameters:

  • inputs (various) – Iterable of light curves / (key, lc) pairs / paths, a glob string, a directory, or a (DataFrame, group_column) tuple.

  • method (str or sequence of str, default "GLS") – Method(s) to run on each light curve.

  • backend (str, default "auto") – Backend selector ("auto"/"cpu"/"gpu"/concrete).

  • device ({"cpu", "gpu", "hybrid"}, default "cpu") – Where to run workers. "hybrid" is not yet implemented.

  • workers (int, optional) – Worker count. None → all-but-one core (CPU) or suggest_gpu_workers() (GPU).

  • grid (GridSpec, optional) – Custom grid applied to every light curve (rare; per-LC defaults are typical).

  • settings (settings model or mapping, optional) – Per-method settings.

  • columns (optional) – Column / domain handling for file and table inputs.

  • domain (optional) – Column / domain handling for file and table inputs.

  • band_column (str, optional) – Multi-band split column for (DataFrame, group_column) inputs.

  • sink (str or Path, optional) – .parquet/.csv file, or a directory (resumable, one part per chunk). None returns rows in memory.

  • n_best (int, default 10) – Peaks stored per light curve.

  • store_raw (bool, default False) – Also store the peak-preserving downsampled spectrum.

  • resume (bool, default True) – Skip chunks already written (directory sink).

  • chunk_size (int, default 256) – Light curves per chunk/task.

Returns:

BatchSummary

Return type:

BatchSummary