Hi guys,
I’m enjoyed a lot of Lesson 2 (listen it for a couple time). I really glad that you start thinking about production side of Machine Learning, and how it is going to be used for end-users.
I’m a web developer, so the production part was a bit confusing for me, so before starting experimenting with my own ideas I decide to test how bear-prediction will work on production.
I was using the most popular web-framework for Python - Django
Here is a very minimalistic version of how it work: http://fastai-bears.lyabah.com/ - upload a picture and get all the prediction.
Here is a source code https://github.com/oduvan/fastai-find-bears feel free to add pull-request, but in education purposes I keep it as simple as possible.
How it works for me:
project bears, the most simple one, contains 2 folders:
- learn - one that you will use on GPU machine
- website - one that you will use on CPU machine
Then, how it works.
- clone the project on GPU machine, close to fastai tutors, and use teach notebook
- after making an export you need to commit a new generated data to git repo and push it on github
- close the project on CPU machine and run in your
docker-compose start
- that’s it.
Happy Coding
PS: Here is the file you might want to edit https://github.com/oduvan/fastai-find-bears/blob/master/bears/website/website/views.py
PPS: If you need I can make a video intro of how it works.