As I just found out the hard way, generating intermediate data dumps is a very time costly action. They are set to occur every n training iterations, but it is not obvious to a new user how fast iterations usually are. Setting too low a value can make the training process take many times as long as without the dumps.
I recommend adding a warning to the app's usage info, or at least a suggestion for what value would be reasonable.