I don't like notebooks.- Joel Grus (Allen Institute for Artificial Intelligence)

I have been using and teaching Python for many years. I wrote a best-selling book about learning data science. And here’s my confession: I don’t like notebooks. (There are dozens of us!) I’ll explain why I find notebooks difficult, show how they frustrate my preferred pedagogy, demonstrate how I prefer to work, and discuss what Jupyter could do to win me over.

Follow O'Reilly on:

[pratikagrawal]

I don't know why I am watching this for 2nd time. This guy is hilarious.

[prabhanshurajpoot]

For people who don't have good speakers. The text editor that the lady (46:27) and other people (49:07) mentioned was
repl.it
Atom extension which allows inline plots and other features:
Hydrogen

[KraziAzian]

I have a programming background so restarting my kernel every and running things top to bottom was always common sense to me. I do agree though that it can be misleading and people may not make that assumption. I think one flaw with his programming style at the end that notebooks solve is having the history of how you checked your data throughout the process. This is really important for cleaning data sets. Unless you save everything you did in iPython along with your python code, it would be hard to trace. For writing models, pure python is definitely better, but for exploring data sets and fixing issues with the data itself, I don't know how notebooks can be beat.

[SY-merk]

I agree with this dude. Unless you are a monk, you will develop bad habits using notebooks. I felt this most when I started out developing code for research and these bad habits wasted a lot of my time.

[jonathansum]

Upvote me if you come here because of Fastai's post.

[akarshjain]

I was intentionally searching for this one. Not just me in the community who thinks it sucks.!!!
This guy did the job....

[NoxmilesDe]

Why was the speed set at 1.5x before uploading?

[markgalassi]

good talk, but the conference organizers need to force people asking questions to wait for a microphone.

[FoleePrime]

I was really at a loss for words when I saw that inline plot at 2:53

[geroldcsendes]

Joel rocks, that is for sure. This made me quit notebooks about 6 months ago and have absolutely no regrets. I brought my data science to the next level. Thanks, Joel!

[lazypunk]

this guy's meme game is on point lol

[richerite]

Very often the datasets we load are quite huge and it's time consuming to just REPL it. This is one advantage of notebooks, where we can annotate intermediate steps with the same dataset

[AlessandroCattabiani]

Joel is giving well thought reasoning to my feeling about notebooks. I always had these feelings but I couldn't point out what or why I did not like them. I totally agree with him

[cswor]

I admit I was fascinated with notebooks when I first encountered them. BUT, I was not developing code that might need to be broken into methods, tested, or placed in production. I was experimenting. Small stuff. Even in that setting, I have found myself migrating to a tool like VS Code. I'm open on what that tool should be except for me, it has to be free. Joel hits a lot of the things that I found bothering me. Extract methods. Intellisense features. Stuff most people who write code for a living will want and benefit from.

Nice talk but he does talk fast.

[OffbrandDrPhil]

No one else saw the loss reference? Just me? Ok, it's at 2:55. Can't miss it.

[SvenTeresniak]

FYI: all the editor (code completion) examples are solved when using JupyterLab with a running language server. But good talk, though, got me thinking.

[RottenRascal]

I am still quite new to data science and I've been using notebooks for my personal projects but listening to this I agree with him and shift my focus. However, I am a little confuse what if I am doing EDA? should I use a notebook or an IDE like Pycharm? because with EDA I am trying to understand the data and that requires me to plot tables and charts as well as code to aggregate data to understand it better. It seems faster to do that in a notebook than to create modules for them especially when viewing graphs isn't as nice in an IDE.

[convincible-uy]

Full agree with this talk. I never understood why a Notebook is go-to tool in Data Science. It is performance limiting tool.
Actually, REPLs did exist for a long time in many languages (Scala, Haskell, Node, Java9, Swift) and they are useful. But we should not abandon IDEs for the sake of in-cell programming.

[wuqui]

It seems to me that Emacs' org-mode and its noweb capabilities could deal with all the problems mentioned?

[shirbenyosef]

be me, start a guide with jupyter, don't like it
use another guide for idle, like it
look for more guide for more knowledge about data science pandas numpy... and all that
the guide uses notebooks.ai
v
(not working looking for advice on youtube and finding this video)
v
the guide actually says in description to use google colab instead cause notebooks.ai isn't working like it used to (and the guide is kinda new)

why, a begginer programmer is expected while learning to use so many tools he does not and cannot fully understand when he absolutely do not need them?

amazing presentation thank you for making me more convinced about learning using simpler tools