Run the CISS-VAE pipeline (letting Python choose device)

Wraps the Python run_cissvae function from the ciss_vae module, handles an optional index_col, and returns imputed data, and optionally the model and silhouette scores.

Usage

run_cissvae(
  data,
  index_col = NULL,
  val_percent = 0.1,
  replacement_value = 0,
  columns_ignore = NULL,
  print_dataset = TRUE,
  clusters = NULL,
  n_clusters = NULL,
  cluster_selection_epsilon = 0.25,
  seed = 42,
  hidden_dims = c(150, 120, 60),
  latent_dim = 15,
  layer_order_enc = c("unshared", "unshared", "unshared"),
  layer_order_dec = c("shared", "shared", "shared"),
  latent_shared = FALSE,
  output_shared = FALSE,
  batch_size = 4000,
  return_model = TRUE,
  epochs = 500,
  initial_lr = 0.01,
  decay_factor = 0.999,
  beta = 0.001,
  max_loops = 100,
  patience = 2,
  epochs_per_loop = NULL,
  initial_lr_refit = NULL,
  decay_factor_refit = NULL,
  beta_refit = NULL,
  verbose = FALSE,
  return_silhouettes = FALSE
)

Arguments

data: A data.frame or matrix (samples × features), may contain NA.
index_col: Character. Column in data to treat as sample ID; removed before training and re-attached. Default NULL.
val_percent: Numeric fraction of non-missing entries to hold out. Default 0.1.
replacement_value: Numeric fill value for masked entries. Default 0.0.
columns_ignore: Character or integer vector of columns to ignore. Default NULL.
print_dataset: Logical; if TRUE, prints dataset summary. Default TRUE.
clusters: Optional vector (or single-column data.frame) of precomputed cluster labels. Default NULL.
n_clusters: Integer for KMeans if clusters is NULL. Default NULL.
cluster_selection_epsilon: Numeric epsilon for HDBSCAN. Default 0.25.
seed: Integer random seed. Default 42.
hidden_dims: Integer vector of hidden layer sizes. Default c(150,120,60).
latent_dim: Integer latent space dimension. Default 15.
layer_order_enc: Character vector for encoder layer sharing. Default c("unshared","unshared","unshared").
layer_order_dec: Character vector for decoder layer sharing. Default c("shared","shared","shared").
latent_shared: Logical; share latent weights? Default FALSE.
output_shared: Logical; share output weights? Default FALSE.
batch_size: Integer batch size. Default 4000.
return_model: Logical; if TRUE, returns Python model. Default TRUE.
epochs: Integer initial training epochs. Default 500.
initial_lr: Numeric initial learning rate. Default 0.01.
decay_factor: Numeric learning rate decay. Default 0.999.
beta: Numeric KL weight. Default 0.001.
max_loops: Integer max refit loops. Default 100.
patience: Integer early stop patience. Default 2.
epochs_per_loop: Integer epochs per refit loop. Default NULL (uses epochs).
initial_lr_refit: Numeric LR for refit loops. Default NULL.
decay_factor_refit: Numeric decay for refit loops. Default NULL.
beta_refit: Numeric KL weight for refit loops. Default NULL.
verbose: Logical; if TRUE, prints progress. Default FALSE.
return_silhouettes: Logical; if TRUE, returns silhouette scores. Default FALSE.

Value

A list with elements:

imputed: data.frame of imputed values (with index_col re-attached).
model: Python VAE object (if return_model = TRUE).
silhouettes: numeric vector (if return_silhouettes = TRUE).