I have not found fully working(install -> create -> save -> use) examples.
I don’t want to reuse legacy code from fastai/old/fastai for loading my new shining models
All viewed by me existed examples have no any tests, requirements.txt/environment.yml and in common don’t looks like solid boilerplate for future experiments
Stack:
flask as simple python server
connection as framework for build API server from Swagger specification. No API boilerplate + auto-generated API doc
And you said it was an incredibly frustrating experience…
I don’t know how can be clearer than that but the v1 shouldn’t be used for the MOOC until the release in January, and the notebooks not advertised because they’re not ready yet.
Well fair, but I got to that point by being incredibly frustrated by the earlier courses not having anything reflected in teh documentation or what people are talking about.
I started doing exactly as prescribed, but I have run into endless dependency hell
@eof as you see, this boilerplate published in " fastai users" (…for help with installing and using the fastai v1 … this library is not used for any fast.ai MOOC) sub-forum.
It’s just one more example for devs who far away from back-end side of Python with best practices, as I understand it . If one want full service - one need to pay. But it’s open-source world You want to make world better - just write your code and make PR
From my experience smoothest one is Ubuntu 16.04 (I need compiled from scratch cv2). I trying 18.04 and latest AWS and can’t recommend both. You need to build from sources too much. With 16.04 i have no any problems on GPU and CPU workstations