I was missing the option to halt training based on a custom condition so I implemented a way to do this using the schedule function.
If the schedule function returns false, training will stop.
This allows you to do a complex early stopping where you stop training if the error rates goes up by a certain amount or doesn't improve beyond a certain amount. It's also minimally invasive to the existing project.
I have updated the documentation to reflect the change.
Thanks for an awesome implementation.
I was missing the option to halt training based on a custom condition so I implemented a way to do this using the schedule function.
If the schedule function returns false, training will stop.
This allows you to do a complex early stopping where you stop training if the error rates goes up by a certain amount or doesn't improve beyond a certain amount. It's also minimally invasive to the existing project.
I have updated the documentation to reflect the change.
I hope you will consider it.