computational efficiency #45
Replies: 4 comments
-
another idea is to use 'label encoding' (not one-hot) for labels and use So last layer is still: |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
when we're ready to work on that, this should be converted to an issue |
Beta Was this translation helpful? Give feedback.
-
I tried implementing some ideas listed here but no joy I also tried to put all the model building commands inside the |
Beta Was this translation helpful? Give feedback.
-
a memory bottleneck in multiclass problems is caused by the use of one-hot encoded label stacks
binary problems can use larger TARGET_SIZE and / or BATCH_SIZE than multiclass problems, for this reason
One possible workaround is to train separate binary models for each class in a multiclass problem
Beta Was this translation helpful? Give feedback.
All reactions