Skip to contents

Plot embeddings via UMAP, colored by arbitrary labels

Source: R/plot_embeddings.R
plot_embeddings_umap.Rd

Computes a 2D UMAP projection of V1..Vd and returns a scatter plot colored by labels membership. Uses cosine distance by default to align with common embedding similarity.

Usage

plot_embeddings_umap(
  embeddings,
  labels,
  n_neighbors = 15,
  min_dist = 0.1,
  metric = "cosine",
  n_epochs = 500,
  seed = 42,
  sample_n = NULL,
  point_size = 2,
  alpha = 0.5
)

Arguments

embeddings

Path to a Parquet file or dataset directory containing columns id and V1..Vd.

labels

Label mapping for ids. Supported formats:

  1. data frame with columns id and label,

  2. path to CSV with columns id and label,

  3. named character vector where names are ids and values are labels,

  4. named list where each element is an id vector for that label.

n_neighbors, min_dist, metric, n_epochs

UMAP parameters passed to uwot::umap(). Defaults: n_neighbors = 15, min_dist = 0.1, metric = "cosine", n_epochs = 500.

seed

Random seed for reproducibility (set to NULL to skip).

sample_n

Optional maximum number of rows to sample for plotting (applied before UMAP). If NULL, uses all rows.

point_size, alpha

Point size and transparency for points in the plot. Defaults point_size = 2, alpha = 0.5.

Value

A ggplot object of UMAP1 vs UMAP2 colored by group.