Maintenance?

[ChungmaruQ]

Training model has to be fixed and advanced during the service.
The number of model will not be changed but indexes could be changed if new classes are reported.

What is the best, and easiest way to maintain the classifier?

[mwang87]

We need to deploy both the model and ontology together.

[bjonnh]

As per #39, would it be possible to keep the numbers in the ontology consistent? Most ontologies these days use fixed numbering for classes that you just increment when you need new classes. That way things that had a specific number in the past will still keep that number. Not sure how you could integrate that with your tensorflow training, but that would be helpful for third-party integration.