Launched: collaboration on Notebooks in Neptune

Hi folks! :wave:


We all love to work with Jupyter notebooks :heart_eyes: Notebooks are standard tool for multiple data science tasks ranging from data exploration to modelling to communicating results. Working interactively and combining code with visuals in a single document is convenient. However, it sometimes generates messy situations, doesn't it?

We are introducing Notebooks space in Neptune. From now on, you can track, share and compare your notebooks in Neptune. Forget about messy notebooks, say hi to notebooks in Neptune :star_struck:

:ringer_planet: Jupyter Extension

You can track your notebook checkpoints directly from Jupter. All you need to do is to install the Jupyter extension. Follow the instructions, then open your notebook and use Configure button. Simple wizard will guide you through configuration. Just paste you API token, select project, then Create Notebook and finally Integrate with Neptune.

That’s it :tada: You are all set and ready to collaborate on Notebooks in Neptune.

configure_notebooks

:spiral_notepad: Notebooks space

We introduce new space in Neptune - Notebooks :partying_face:
All your Notebooks can be tracked, each can have as many checkpoints as you wish:

  • Review who created latest checkpoint (and when),
  • Analyze notebook directly in Neptune or download it and continue working on it,
  • Use Upload button to upload new checkpoint to Neptune,
  • Share your results and insights with others by sharing link.

com-video-to-gif

:rocket: Now imagine…

That you can compare everything in Notebooks side-by-side like regular Python code. Compare cells (code or markdown), outputs, visualizations, everything!
Stay tuned, this is coming :sunglasses:

:mag_right: Did you know…

…you can track your MLflow experiments and still take advantage of collaboration in your projects? Read more here!


Do you like new Notebooks space and Jupyter extension?
We are always open for comments and feedback - contact us!

Cheers!
Paulina

Thank you, it looks like it is a great feature! Just installed neptune.ml extension to my Jupyter. What is the right workflow for using notebooks and experiments at same ML project? Is there any relationship between notebook version and experiment results?

Hi @Maxyz, great to see you here!

I am working on an example project with notebooks and experiments in the same project.
You can check it out here.

Basically, I work on exploratory analysis, visualizations, and feature engineering prototyping in my notebooks.
Once I am done I usually run a .py script which create experiments.
I combine notebooks and experiments to communicate my work in the wiki workspace.

That being said, you can defiinitely version notebooks (they will be stored in the notebooks section in the app) and run model training in notebooks (it will be saved to the experiments section). If you want to save snapshots of .ipynb while experimenting you should use:

with neptune.create_experiment(upload_source_code=['train_model.ipynb']):
   ...
   logic
   ...

it will save the .ipynb snapshot and attach it to the experiment in the Source code section.

We really want to be super flexible and let people do it the way they want to. That means there is no standard here but I like to explore/visualize/prototype in notebooks and run feature engineering/training/tuning in the experiments.

I hope this helps!

1 Like