Using #563 I am trying to fix GitHub CI jobs that seem to fail in minibatch limits
Found that the latest TensorFlow release doesn't like using concat on scalars so this has been changed to stack.
The Einstein averaging for remainder case was also wrong but this is fixed now and local testing demonstrated that it works.
I could test the atomwise batching and it worked great!
We should consider how best to add low memory cases. We can run all of the tests in a memory_fraction=0.9 and then a memory_fraction=0.01 and that should test all cases. @christophlohrmann @PythonFZ what do you think? It should probably come in a different PR once we decide on the best format for duplicating tests only with one configuration parameter changing
Using #563 I am trying to fix GitHub CI jobs that seem to fail in minibatch limits
We should consider how best to add low memory cases. We can run all of the tests in a memory_fraction=0.9 and then a memory_fraction=0.01 and that should test all cases. @christophlohrmann @PythonFZ what do you think? It should probably come in a different PR once we decide on the best format for duplicating tests only with one configuration parameter changing