peal.training.loggers¶
TensorBoard logging for predictor (and generator) training.
Logger is owned by ModelTrainer: it accumulates the per-step losses
and predictions of one epoch, writes scalar / histogram summaries with the
0_ (validation), 1_ (train), 2_ (per-attribute) and 3_
(everything else) tag prefixes that order the TensorBoard panels, and
returns the epoch metric early stopping is judged by.
log_images_to_writer dumps a few sample batches of a dataloader as image
grids so a run’s inputs can be inspected in TensorBoard.
Functions
Write up to three sample batches of a dataloader to TensorBoard. |
Classes
Accumulate step results of one epoch and write TensorBoard summaries. |