Closed douglasbagnall closed 2 years ago
However, I just noticed the CI tests failed when I rebased on upstream
https://github.com/catalyst/moodle-mlbackend-python/runs/2680324813
which I think might be down to Tensorflow versions.
So don't merge yet!
Let's close this one. Some of the changes have ended up in #51; others will probably not apply cleanly after that lands. None of it is urgent.
I'll open a new PR in due course.
These are a few fixes that have been running in production for a year or so.
Better control of where temporary files go and when they disappear.
We don't always want to write in
/tmp
which might be on a small partition or something, and we don't want to leave around files once we don't need them.Allow version masquerading
From an API perspective, all versions from 2.3 to 3.1 are compatible (at least from the point of view of the actually used API). This means that although an ageing Moodle might claim it wants version 2.4, it could actually cope with version 3.1, if it took the trouble to retrain its models (the stored models are not compatible). We allow administrators who know what they are doing to lie to Moodle about the true version without hacking, enabling them to have a newer version of the backend.
(Note, there should really be three versions: one for the API, one for the on-disk model format, and one for the code, and Moodle should only care about the first two).
Miscellaneous fixes