Thrilled to discover nbdev from & . It's an attempt to solve a big problem with computational notebooks like Jupyter: you explore problems with a notebook, but usually need to "switch" to a more powerful tool for "real" impls:
Conversation
nbdev tries to solve this problem by giving you
- automatically turning notebooks into publishable Python modules
- bidirectional sync with plaintext .py for IDE usage
- fixes for other "real" project needs: tests, continuous integration, documentation export, conflict resolution
1
2
19
While nbdev seems focused on helping individual developers avoid the "switch" when implementing their own projects, I think layers like this could help solve a big problem with "executable books": the huge barrier for *readers* to build on embedded code to do anything real.
1
9
The fast.ai docs get at something pretty exciting, then. Like many notebooks, it contains narrative content which explains computational material and lets readers explore. But the executable book is *also* the implementation of a published production-level library
1
14
The narrative in those docs is a bit limited: it's more documentation than prose. "Deep Learning for Coders" is the expository text from the same authors, but it isn't made available in an executable context AFAICT. I think that could be really powerful!
Replying to
(For more on these themes and on dreams of executable books, see numinous.productions/ttft/#executab)
2
3
22
My wishlist for exec. books:
– author did their "real thinking" in the authoring computational environment
– reading environment invites + supports meaningful experimentation/exploration
– book elements transparently and usefully reusable by author and readers in derivative works
4
16
Update: I was wrong; "Deep Learning for Coders" is in fact available in notebook format! Most excellent!
Quote Tweet
Replying to @andy_matuschak
The book is in fact available in executable format, for free! :D
github.com/fastai/fastbook
2
25