JupyterLab adoption vs Classic Notebook

JupyterLab
1

Out of curiosity, I ran three polls last week on Twitter, LinkedIn, and YouTube asking the following question:

Jupyter users only: Do you use Jupyter Notebook or JupyterLab these days?

The answer choices I offered were:

  • Jupyter Notebook
  • JupyterLab
  • Both
  • Just show me the results

Ignoring the final category, the polls received a total of around 2666 votes, which broke down as follows:

  • Jupyter Notebook: 62%
  • JupyterLab: 30%
  • Both: 8%

I have a couple questions for the forum about these results:

  1. If you interact with a lot of Jupyter users, do these results track with your own experience? (Although it was a large-ish sample, it’s not necessarily a representative sample of users.)
  2. How does this compare with the “expected” pace of JupyterLab adoption?
  3. Does the pace of JupyterLab adoption matter? Meaning as a community, is it important that we actively encourage more and more users to move to JupyterLab?

For what it’s worth, my take is that most people start with Jupyter Notebook and either aren’t aware of JupyterLab or don’t see a reason to switch to it. And for those folks who want a more full-featured editor, many prefer using VS Code with a Jupyter extension (and arguably, I should have listed that as a poll option).

In fact, I’m a long-time Notebook user, and I know I could benefit from Lab, but I just haven’t taken the time to switch! Plus I often open gigantic notebooks, and if there’s any additional lag in Lab (compared to Notebook), that would be a significant disadvantage for me.

2

Was it clear that “Jupyter Notebook” refers to the application, and not the format/ecosystem in general?

1 Like
3

Excellent question!

No, I didn’t add any additional explanation, though I would assume (perhaps incorrectly) that if you’re a JupyterLab user, you know that you’re using JupyterLab and thus would select that poll option.

In other words, I generally wouldn’t expect that a JupyterLab user would click the “Jupyter Notebook” option. But you’re right, there are probably people who use JupyterLab and don’t know the name of the interface, but they do know that they use “Jupyter notebooks.”

2 Likes
4

Hi,

dataschool
February 20

Excellent question!

No, I didn’t add any additional explanation, though I would assume (perhaps incorrectly) that if you’re a JupyterLab user, you know that you’re using JupyterLab and thus would select that poll option.

In other words, I generally wouldn’t expect that a JupyterLab user would click the “Jupyter Notebook” option. But you’re right, there are probably people who use JupyterLab and don’t know the name of the interface, but they do know that they use “Jupyter notebooks.”

I think the idea is that there may be people who have ended up using
JupyterLab without ever using the classic notebook interface, and they
would think of that as using the “Jupyter Notebook”.

There is some long discussions about Classic vs JupyterLab, including
various survey results, here:

Cheers,

Matthew

3 Likes
5

Thank you for the links! I look forward to checking them out.

6

Thanks for sharing the results of your poll. We can also check Pypi download stats for “notebook” (PyPI Download Stats) and “jupyterlab” (PyPI Download Stats). Counting each download during the last month as a “vote”, we have:

>>> notebook = 11526130; lab=2056256
>>> notebook/(notebook + lab)                          
0.8486086317970937
>>> lab/(notebook + lab)     
0.15139136820290633
2 Likes
7

Download numbers are always funny without creepy telemetry: without unique IPs, and weeding out CI machines, even the relative numbers can be skewed, but 80/20 is not surprising.

Another wrinkle: every install of jupyterlab also incurs a download of notebook, even if it’s not being actively used, bringing it closer to the old rule-of-thumb of 80/20.

Further, the (troublesome) jupyter metapackage triggers a download of notebook and not of jupyterlab… but does bring in qt :crying_cat_face:, which is kind of a branding fail.

After Notebook 7 is released, the dependency will be reversed (as it will be built on lab), so these relationships will get even murkier.

3 Likes