Open drop-ella opened 5 years ago
I modified spotlight implicit classes by adding notifier function for every epoch completion, so you have full control over however you can do with loss and self._net model. I use to write loss and my own eval function in tensorboard and works great.
Also I did same thing for writing batch loss at every 500 batches.
Tensorboard is now supported by PyTorch. Does spotlight support it or will do so in the future? And what is a suggested way of monitoring model training (training and validation loss/metrics at step n) when using spotlight models?
Thank you very much!