Set renderer props. Returns true if a render is needed.

Effective scale of embedded points. Default: 1.0.

Advance the layout by `n_epochs` at the current learning rate (no decay). Intended for interactive/real-time stepping.

Advance the layout by `n_epochs` at the current learning rate (no decay) — intended for interactive/real-time stepping.

Test audio embedding with CLAP via transformers.

Test that auto modality correctly detects audio.

Test that backoff from one task's failure delays other concurrent tasks.

Test that a successful call resets the shared backoff so later calls aren't delayed.

Test that errors cause subsequent calls to be delayed via shared backoff.

Should produce correct number of edges.

Test basic async mapping with a simple function.

Bipartite graph: group A links to group B, B links back to A.

Should add a 'pagerank' column to the DataFrame.

Test that concurrency limit is respected.

Test custom column names for x, y, and neighbors.

Verify the custom embedder receives canonical image dicts.

Verify model and embedder_args are forwarded to the custom embedder.

Verify the custom embedder receives canonical text (list[str]).

damping=1.0 means no teleportation: pure link-following.

damping=1.0 on asymmetric graph follows link structure only.

damping=0 means pure teleportation: uniform distribution.

A node with no outgoing edges (dangling) should still get rank.

Test async_map where return type differs from input type.

Test that processing continues when fallback is provided even on errors.

Duplicate edges should be coalesced (weights summed) by sparse tensor.

UMAP-style weights should be positive.

Our edge weights should match UMAP's compute_membership_strengths.

Empty edge list with n=0 should return an empty array.

Empty edge list with explicit n should return uniform scores.

Test async_map with empty input list.

Test that fallback value is used when function fails and fallback is provided.

Test that fallback is used after all retries are exhausted.