Make ModelCheckpoint backwards compatible with TF <1.14
I attempt to create a ModelCheckpoint callback with the following code:
model_checkpoint = ModelCheckpoint( HOME + "/locust/best_model_SHG.h5", monitor="loss", #monitor="loss" # use if validation_split=0 verbose=1, save_best_only=True, )
Upon running it, the code fails with the following error:
Traceback (most recent call last):
File "R:/Locust/bc175/Warren Research Technician/deepposekit/Training.py", line 147, in
When I remove that code, training works fine, so it's definitely what's causing the issue.
Thanks, Ben
Thanks for pointing this out. What version of TF are you using? It seems the ModelCheckpoint callback will only work with TF >=1.14. If you're using a version <1.14 I'd recommend upgrading to a newer version. When I get the chance I will update so it's backward compatible.
Thanks - I was using 1.13, upgrading to 2.0.0 sorted the problem!
Glad it's working. I'm going to reopen this so I don't forget about it.