KexinFeng

If NAN happens only in the middle of the training not the beginning, then it is a little troublesome to debug. In this case, I would print the gradient of...

Can we simply add a zero_grad() step in the PyTorch implementation of backward call @siddvenk ? This solution assumes that repetitive calling of zero_grad() is ok, considering in Trainer there...

Could you be more specific about what binary image classification task you are looking for? You mentioned "detect if a Image is similiar to another one", it sounds a lot...

@DxsSucuk You can consider using transfer learning; use pre-trained model to get vector embedding of pictures. Then it can be used for the downstream tasks like one-class classification. We are...

@patins1 It looks like `BulkDataIterable` is not covered by any unit test. You mentioned that using `BulkDataIterable` is more efficient. Could you add a unit test that covers this class...

> The problems with MxNet is fixed by using "take" operation instead of "pick". > TrainMnist executes now 4 times faster for MxNet . Could you integrate it with the...

@patins1 Thanks for dealing with this issue in such a timely manner! But it looks like the unittest that covers the new file api/src/main/java/ai/djl/training/dataset/BulkDataIterable.java is still missing. You mentioned you...

@siddvenk Hi Siddarth, in this pr, we have changed the definition of get(NDArray index) from pick to take, and have given the warning. This will affect the result you mentioned...

@r1ckypp We have just added the support for gluonTS model inference. Check out https://github.com/deepjavalibrary/djl/pull/1903

@mxnet-bot run ci [unix-cpu, unix-gpu, windows-gpu]