I would appreciate feedback regarding an issue I am having with the CrossEntropyLossLayer[“Index”] function. I have attached a notebook with detail explanation of the issue. I have run the code before and it run well. It now requires to adjust the labels of the MNIST dataset between 1 and 10, instead of the 0 to 9 in the original dataset. Is there is a way around so that no additional adjustment of the labels is needed?. In the past the adjustment was not required.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Documentation says we can provide a loss layer to NetTrain as 3rd argument, why doesn't this work? | 2 | 1911 | September 23, 2025 | |
| Correct format for NetChain? | 6 | 15701 | January 22, 2018 | |
| Create a custom loss function with NetTrain? | 3 | 13475 | August 14, 2018 | |
| SoftmaxLayer with one-hot vectors | 2 | 4278 | May 14, 2024 | |
| Add a customized complex loss function to a neural network? | 3 | 6803 | February 6, 2019 |