AnnData functions

Need the anndata extra. See AnnData / scanpy.

umap

umap(
    adata: AnnData,
    *,
    n_components: int = 2,
    min_dist: float = 0.5,
    spread: float = 1.0,
    neighbors_key: str = "neighbors",
    key_added: str | None = None,
    device: str = "cpu",
    seed: int = 42,
    verbose: int = 0,
    copy: bool = False,
    **kwargs: Any,
) -> AnnData | None

UMAP on the sc.pp.neighbors graph, stored the way sc.tl.umap does.

The neighbour graph is taken from obsp rather than searched again, so n_neighbors and the metric are whatever sc.pp.neighbors used. The representation (use_rep / n_pcs) is read from the same uns[neighbors_key]["params"]; it feeds the initialisation.

Parameters:

Name	Type	Description	Default
adata	AnnData	Annotated data matrix, after sc.pp.neighbors.	required
n_components	int	Output dimensionality.	2
min_dist	float	How tightly points may pack. Fits the repulsion curve with spread.	0.5
spread	float	Scale of the embedding relative to min_dist.	1.0
neighbors_key	str	Where sc.pp.neighbors stored its results in uns.	'neighbors'
key_added	str | None	Store under obsm[key_added] and uns[key_added] instead of obsm["X_umap"] and uns["umap"].	None
device	str	"cpu" for manifolds_rs.UMAP, "gpu" for manifolds_rs.UMAPGpu.	'cpu'
seed	int	Fixes the initialisation and the negative sampling.	42
verbose	int	0 silent, 1 normal, 2 detailed.	0
copy	bool	Return a modified copy instead of writing into adata.	False
**kwargs	Any	Any other constructor argument of the chosen estimator, e.g. n_epochs, init, optim_params. Unknown names raise.	{}

Returns:

Type	Description
AnnData | None	The copy if copy, else None. Sets obsm["X_umap" | key_added]
AnnData | None	and uns["umap" | key_added]["params"].

Raises:

Type	Description
ValueError	If sc.pp.neighbors has not run, its graph is ragged, or the representation is missing or sparse.

tsne

tsne(
    adata: AnnData,
    n_pcs: int | None = None,
    *,
    use_rep: str | None = None,
    perplexity: float = 30.0,
    approx: str = "barnes_hut",
    key_added: str | None = None,
    device: str = "cpu",
    seed: int = 42,
    verbose: int = 0,
    copy: bool = False,
    **kwargs: Any,
) -> AnnData | None

t-SNE on a representation, stored the way sc.tl.tsne does.

Searches its own 3 * perplexity neighbours; sc.pp.neighbors is not used.

Parameters:

Name	Type	Description	Default
adata	AnnData	Annotated data matrix.	required
n_pcs	int | None	Leading columns of the representation to use. None uses all.	None
use_rep	str | None	None for obsm["X_pca"], "X" for adata.X, or any other obsm key. Must be dense.	None
perplexity	float	Effective neighbourhood size.	30.0
approx	str	"barnes_hut", "qd", "fft", "fft_3k", or "fft_3k_gpu" with device="gpu".	'barnes_hut'
key_added	str | None	Store under obsm[key_added] and uns[key_added] instead of obsm["X_tsne"] and uns["tsne"].	None
device	str	"cpu" for manifolds_rs.TSNE, "gpu" for manifolds_rs.TSNEGpu.	'cpu'
seed	int	Fixes the initialisation.	42
verbose	int	0 silent, 1 normal, 2 detailed.	0
copy	bool	Return a modified copy instead of writing into adata.	False
**kwargs	Any	Any other constructor argument of the chosen estimator, e.g. n_epochs, learning_rate, ann, metric. Unknown names raise.	{}

Returns:

Type	Description
AnnData | None	The copy if copy, else None. Sets obsm["X_tsne" | key_added]
AnnData | None	and uns["tsne" | key_added]["params"].

Raises:

Type	Description
ValueError	If the representation is missing or sparse, or approx is not available on device.