Our team is into deep learning and uses kerasmxnet (a really old and deprecated package: https://github.com/awslabs/keras-apache-mxnet). The package is today limiting us to Python 3.6 and I am trying to upgrade us to 3.7 or even 3.8 if possible. I’m curious as to whether or not this package works with Python 3.7 (or even 3.8). My rationale is that it should given Python 3.7 and 3.8 are backwards compatible (ie any Python 3.6 code should run fine on 3.7 and 3.8).
Questions:
Is my rationale correct? Can I extend it and apply to KerasMXNet package?
FWIW, I have been able to upgrade to 3.7 and 3.8 fine and verified that our unit tests and integration tests (on CPU) work fine. However I’m not sure about behavioral tests (which will take some time to run) and that’s why this question to short circuit if there’s knowledge available that says “don’t do it, this isn’t recommended”.
Note: we are planning to migrate to PyTorch eventually but this is a bigger migration and I’m trying to incrementally move to Python 3.8 first before taking on the migration to PyTorch.
