With these changes, it is possible to configure when weights are saved turning training. Per default, weights are stored every 1000 iterations, if max_batchsize is less than 10000, or every 10000 iterations otherwise. For smaller datasets, it is useful to store them at different rates.
With the proposed code changes it is possible to read these settings from the data config file.
For example with the following config weights will be stored every 2000 iterations (*_2000.weights, *_4000.weights, ...) and the *_last.weights will be stored after every 500 iterations in the backup folder.

classes = 80
names   = coco.names
saveweights = 2000
savelast 	= 500

an exemplary config.data file