Closed colibrisson closed 1 year ago
Are you sure you don't want to log the validation metric as well? It is used for certain decisions like the early stopping and the learning rate reduction on plateau so it might be useful to log it in addition to the raw accuracy.
In addition I would like to add it to the segmentation and pretraining as well.
Add Tensorboard support as proposed in mittagessen#422.
Tensorboard logging is triggered by adding
--debug
flag to theketos train
command. The metrics logged are training loss, validation accuracy, and learning rate.