Compute a group-level embedding from a triplet data list

Trains a single embedding on the combined triplet judgments from all participants. Individual per-participant embeddings are not computed. Use this function when you only need a group summary of the similarity structure, which is faster than run_embeddings_from_list.

Usage

run_group_embedding_from_list(
  triplet_list,
  d = 5L,
  max_epochs = 50000L,
  tolerance = 1e-04,
  tol_window = 10000L,
  seed = 222L,
  device = NULL
)

Arguments

triplet_list

A named list of data frames, one per participant, as returned by get.combined. Each data frame must contain the columns Center, Left, Right, Answer, sampleAlg, and sampleSet.

d

Number of embedding dimensions. Default 5.

max_epochs

Maximum number of training epochs. Default 50000.

tolerance

Loss tolerance for early stopping. Default 1e-4.

tol_window

Epochs without improvement before early stopping triggers. Default 10000.

seed

Integer random seed for reproducibility. Default 222.

device

PyTorch device string, or NULL (default) to auto-select: CUDA GPU if available, then Apple MPS, then CPU. Pass "cpu" to force CPU even on a GPU machine.

Value

A named list with three elements:

embedding

Numeric matrix with one row per item (item names as row names) and d columns (dim_0, dim_1, …).

loss

Best test loss achieved during training.

history

Data frame with one row per epoch and columns epoch, train_loss, test_loss, train_acc, test_acc.

Item indexing

All unique item names appearing in the Center, Left, and Right columns across all participants are collected and sorted alphabetically. Each item's zero-based index in this sorted list is used as the integer index for the Python model.

Filtering

Trials with sampleAlg == "check" are excluded. The sampleSet column ("train" / "test") must be present and is used to split data for early stopping.

Examples

if (FALSE) { # \dontrun{
grp <- run_group_embedding_from_list(
  triplet_list = icon_triplets,
  d            = 3L,
  max_epochs   = 50000L
)

# Embedding matrix (items x dimensions)
head(grp$embedding)

# Best test loss
grp$loss
} # }