Skip to contents

Rarely needed knobs of the UMAP algorithm. Pass the result to the control argument of stat_umap(), geom_umap() or umap_embed().

Usage

umap_control(
  learning_rate = 1,
  negative_sample_rate = 5L,
  repulsion_strength = 1,
  local_connectivity = 1,
  set_op_mix_ratio = 1,
  a = NULL,
  b = NULL,
  optimizer = NULL,
  disconnection_distance = NULL,
  progress = getOption("ggumap.progress", TRUE)
)

Arguments

learning_rate

Initial learning rate of the layout optimisation.

negative_sample_rate

Number of negative samples per positive sample.

repulsion_strength

Weight of negative samples (gamma in umap-learn).

local_connectivity

Number of nearest neighbours assumed to be connected at a local level.

set_op_mix_ratio

Mix between fuzzy union (1) and fuzzy intersection (0) when symmetrising the neighbour graph.

a, b

Parameters of the low-dimensional similarity curve. NULL (default) derives them from min_dist and spread.

optimizer

NULL (default for the chosen flavour), "sgd", "momentum" or "adam". Momentum and Adam require flavour = "0.6".

disconnection_distance

Distance at or above which points are considered disconnected. NULL uses the metric's default.

progress

Show a progress bar for long computations? Defaults to the option ggumap.progress (TRUE if unset). Bars appear only after a couple of seconds.

Value

A list of class ggumap_control.

Examples

umap_control(negative_sample_rate = 10)
#> <ggumap_control>
#> learning_rate: 1
#> negative_sample_rate: 10
#> repulsion_strength: 1
#> local_connectivity: 1
#> set_op_mix_ratio: 1
#> a: NULL
#> b: NULL
#> optimizer: NULL
#> disconnection_distance: NULL
#> progress: TRUE