Flux is Improved with Zygote
I highly recommend that you don’t give up on Flux. The use of Zygote was always planned and actually makes Flux much better. I am not sure when I will get around to updating my article, but it is based mostly on the official Flux intro code examples which have in fact been updated.
I have also written a new article based on the new and improved Flux, but covering handwriting recognition which you could also look at.
The benefits of Zygote is we get rid of the tracked numbers (or dual numbers) because Zygote is able to transform the code for you when the JIT runs.
I have started writing a Machine Learning book using Julia and Flux. But it is in fits and starts. I hope to make some of it available on my blog site. That is one reason I may not update this article. I just find that the Machine Learning topic is too large to really write a single Medium article.
It is much better to have a site where you can organize articles and cover several topics. If you look at my handwriting article you can see I have listed several related articles. I am trying to hash out different topics which needs to be covered together.
A beginner getting into Machine Learning need to be made familiar with basic linear algebra, julia references, differentiation, dual numbers, how automatic differentiation works. How Flux works, maybe how it is different from things like TensorFlow.
Currently some of my challenge is picking cool examples to build the text around. I want some fun examples to have code examples for which gets people engaged.
Suggestions for topics and example code is very welcome!