mantispy.pp.downsample#

mantispy.pp.downsample(adata, n_per_group=500, by=('Metadata_Plate', 'Metadata_Well'), stratify=None, seed=0)[source]#

Return at most n_per_group rows from each group.

Parameters:

adata AnnData: Object to sample from. Never modified.
n_per_group int (default: 500): Cap per group. Groups smaller than this are kept whole, so groups are capped but not balanced.
by Sequence[str] | str | None (default: ('Metadata_Plate', 'Metadata_Well')): Columns defining a group. The default caps each well, so every well is represented instead of the densest wells filling the sample.
stratify str | None (default: None): Keep this column’s proportions inside each group, so a rare perturbation is not lost to the sampling.
seed int (default: 0): Seed for reproducibility.

Return type:

AnnData

Returns:

A new object holding the sampled rows in their original order, with the call recorded in uns["mantispy"]["params"].

Raises:

ValueError – If n_per_group is below 1.