Create a training logger for factor network fitting

Returns a logger object that records per-iteration metrics during fit(). After training, the log can be printed, plotted, or exported to CSV.

Usage

training_logger(
  log_loss = TRUE,
  log_norms = FALSE,
  log_classifier = NULL,
  interval = 1L
)

Arguments

log_loss

Log reconstruction loss per layer (default TRUE).

log_norms

Log factor W/H norms per layer (default FALSE).

log_classifier

Evaluate classifier accuracy per iteration using a supplied classify_embedding configuration. Must be a list with labels and optionally test_idx, k, side, and layer (default NULL = disabled).

interval

Log every interval iterations (default 1).

Value

A training_logger object to pass to fit().

See also

nmf, fit

Examples

if (FALSE) { # \dontrun{
logger <- training_logger(
  log_norms = TRUE,
  log_classifier = list(labels = labels, test_idx = test_idx, k = 5)
)
res <- fit(net, logger = logger)
print(logger)
plot(logger)
export_log(logger, "training_log.csv")
} # }