A blog post I published on TowardsDataScience.com, presenting a use case of the Keras API in which resuming a training process from a loaded checkpoint needs to be handled differently than usual. Includes examples and a tutorial on how to solve this issue.