Closely related to:
https://tensorflow-prod.ospodiscourse.com/t/model-checkpointing-best-practices-when-using-train-step
When I am using this the callback is unable to keep track of the metric values correctly:
I notice something similar in the SimSiam example I linked in my previous post as well. Am I missing out on?
Are these metrics outputs both before or after the weight update step?
Cause if not these printed values are not the same
Also have you implemted def metrics(self) or called reset_states() if you have used a custom model with a custom train loop?
Refer to this example: Self-supervised contrastive learning with SimSiam. It’s all laid out there.
@Bhack you are actually right. I probably need to implement a tracker for this to work. I will do that and update here. Sorry.
