univi.datasets.export_dataset

Contents

univi.datasets.export_dataset#

univi.datasets.export_dataset(name, adatas, outdir, *, paired=None, labels=(), obs_columns=None, splits=None, split_key='split', counts_layer='counts', title=None, description=None, source=None, compression='gzip', overwrite=False)[source]#

Write AnnData objects as a hostable dataset; return the manifest path.

Parameters:

  • name – Dataset name as it will appear in univi.datasets (e.g. "hao_citeseq_pbmc").

  • adatas – {key: AnnData}, e.g. {"rna": rna, "adt": adt}. Objects are not modified.

  • outdir – Output directory (created).

  • paired – Keys that describe the same cells. They must contain the same cell ids; they are written in the order of the first key. Default: all keys when they share identical cell ids, otherwise none.

  • labels – Label columns expected in .obs (checked and recorded in the manifest).

  • obs_columns – .obs columns to keep (default: all). split_key is always added when splits is given.

  • splits – {"train": ids, "val": ids, "test": ids} with cell ids or positional indices into the first paired object. Stored as obs[split_key] ("unassigned" elsewhere) in every paired object.

  • counts_layer – Layer holding raw counts; .X is used when the layer is missing.

  • title – Text for the manifest and DESCRIPTION.md.

  • description – Text for the manifest and DESCRIPTION.md.

  • source – Text for the manifest and DESCRIPTION.md.

Contents