Projected epoch count for the training progress bar

The run stops when patience epochs pass without a better validation loss, so if the current best never improves it ends at best_epoch + patience. That is the projection; restarts not yet started are budgeted at the mean length of the ones already finished.

Usage

train_progress_total(
  epochs_done,
  best_epoch,
  patience,
  max_epochs,
  restart,
  n_restarts
)