Neural Network Checkpoints only every 'N' Epochs??
2 次查看(过去 30 天)
显示 更早的评论
In the MATLAB documentation for Deep Learning Toolbox (formerly Neural Network Toolbox) we see that we can use trainingOptions with a 'CheckpointPath' to save checkpoints during training.
While better than no checkpoints, this doesn't seem very flexible since it forces one to save either no checkpoints, or save all checkpoints after every epoch. In problems where an epoch takes a long time, this is fine, but a lot of problems rip through epochs and train for 10k or even 100k+ epochs, and I would imagine I'm not the only one who would like to save only every Nth checkpoint on problems like these.
Is there a workaround that isn't a complete hack to write every Nth checkpoint? The only thing I can think of is to use the 'OutputFcn' option in trainingOptions, but you still will write every checkpoint and then you use the OutputFcn to just delete what was just saved except for every Nth time, where you just let the OutputFcn return false. Seems like there should be a better way though...
0 个评论
回答(0 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!