Wrapper function to generate parametric UMAP parameters

Wrapper function to generate parametric UMAP parameters

Usage

params_parametric_umap(
  local_connectivity = 1,
  bandwidth = 1e-05,
  mix_weight = 1,
  hidden_layers = c(128L, 64L, 32L),
  lr = 0.001,
  corr_weight = 0,
  n_epochs = 500L,
  batch_size = 256L,
  neg_sample_rate = 5L
)

Arguments

local_connectivity

Numeric. Number of nearest neighbours assumed to be at distance zero. Defaults to 1.0.

bandwidth

Numeric. Convergence tolerance for smooth kNN distance binary search. Defaults to 1e-5.

mix_weight

Numeric. Balance between fuzzy union and directed graph during symmetrisation. Defaults to 1.0.

hidden_layers

Integer vector. Hidden layer sizes for the MLP encoder. Defaults to c(128L, 64L, 32L).

lr

Numeric. Learning rate for the neural network optimiser. Defaults to 0.001.

corr_weight

Numeric. Coefficient for the negative Pearson correlation loss that encourages similar distances in embedding and original space. Defaults to 0.0.

n_epochs

Integer. Number of training epochs. Defaults to 500L.

batch_size

Integer. Training batch size. Defaults to 256L.

neg_sample_rate

Integer. Number of negative samples per positive edge. Defaults to 5L.

Value

A list with the parametric UMAP parameters.