Writing history.json after each epoch
In these lines the history.json
file is written out after the full training, which will not happen in case the job fails, times out etc.
We used to write this out after each epoch, we should go back to this syntax and add it into a callback.
Also the syntax between the validation and the training history output is not consistent, e.g. one has lr
as key the other learning_rate
Edited by Manuel Guth