A crap deep learning library. Although when I say crap I just mean not as good as fast.ai, but probably still better than TensorFlow :-P
Part 2 of the Fast.ai course (2019) is about recreating much of Fast.ai from the ground up.
The end result is a library you can experiment with if you think you might have a better way of doing things, or somewhere you can try implementing academic papers which you want to play around with.
I want to understand the basics of a deep learning library well enough to be able to put one together and tinker with it. That's just how my brain works; I was 'that kid' who took the radio apart at 7 years old to try and understand how it worked.
This isn't intended to ever be a production ready lib like Fast.ai, but somewhere I can learn to better understand how to get the most out of production ready libs like Fast.ai, and somewhere to play around with new ideas from time to time.
Once this eventually gets checked in, it still may not be that interesting to look at. Anything worth discussing will end up as blog posts and/or contributions to Fast.ai.
pip install jmd_crapnet
Fill me in please! Don't forget code examples:
1+12
hello()crap