5 ms·

Show HN: Render Jupyter notebooks as interactive articles

217 points by adlha 5y ago

adlha 5y ago
Hey everyone, our team just built a tool to make your Github notebooks interactive and we would love for you to try it out. Just enter the GitHub URL of your (or any) public notebook and hit Render to instantly view the notebook. Your notebook will be rendered in an article-like layout and will get a table of contents. Anyone can fork it using the "Launch" button and play around with your code. The Deepnote viewer is also faster and more reliable than other .ipynb viewers we've tried. Would love for you to take it for a spin & hear your feedback!
- NmAmDa 5y ago
  That seems great, but for the comparison between DN viewer and the other, is there any benchmark tests specifically or it just a general feeling of speed when someone is using them. Because for each Jupyter notebook viewer it will depend on the setup, machine and the load itself.
  - the21st 5y ago
    Hi there! Simon (Engineer at Deepnote) here. We didn't benchmark the load speed but for some reason, Github's ipynb viewer has always felt to me quite slow and unreliable. All viewers are publicly accessible so I'd love it someone did an independent benchmark. I'd prefer to avoid doing one ourselves because of the obvious conflict of interest
    - verdverm 5y ago
      Why not make a benchmark which is also open and reproducible?
    - NmAmDa 5y ago
      I thought that it is more than this. I think you should support any claim about that by creating a reproducible open benchmark test so at least people can understand what criteria you selected.
      - NmAmDa 5y ago
        I also find it myself much faster than GitHub viewer.
- erikreinertsen 5y ago
  I am a huge fan of Deepnote! Have you considered / is there already a way to “import” an .ipynb into a DeepNote project? Would be easier for users than to copy and paste code one cell at a time.
  - adlha 5y ago
    Absolutely! All you have to do is Upload your .ipynb files (or full folders) into an empty Deepnote project and run the code. I'm sending docs with more info on how to import files here: https://docs.deepnote.com/importing-and-exporing/importing-data-to-deepnote https://docs.deepnote.com/importing-and-exporing/importing-d...
- carreau 5y ago
  What's your security model for logged-in users, one of the reason we (the jupyter team) have nbviewer on a separate domain with no login is to have embedded JS and other potentially sensitive content to be rendered without risk. We've seen people trying many attack vectors against renderer like this one with for example injecting script tags in things like prompt numbers.
  - sunaden 5y ago
    Thanks for the q (I work at Deepnote) - all outputs that can contain potentially malicious JS are sandboxed in iframes so they can only access their local context and can't be used e.g. for XSS attacks.
codeknight11 5y ago
I've been using Deepnote for a while and highly recommend it. I love the features and they have also built a great community. Github's jupyter rendering doesn't work for me most of the time so I am using Deepnote viewer these days. Check out my published article that I wrote using Deepnote a while ago - https://deepnote.com/@tj/The-Starry-Cat-7HRaWnp8QhWYeK_x_QLD9Q https://deepnote.com/@tj/The-Starry-Cat-7HRaWnp8QhWYeK_x_QLD...
dmytton 5y ago
I've been using Deepnote to do some energy modelling for an academic paper I'm writing, and it's been great. The UI is much nicer than any other Jupyter editor I've used, and the integration with GitHub works well. Just started collaborating with some other researchers on another project so the real-time editing is coming in handy. So much better than using Excel. I also interviewed their CTO, Jan Matas, as part of my developer interview services about how it's all built around k8s behind the scenes: https://console.dev/interviews/deepnote-jan-matas/ https://console.dev/interviews/deepnote-jan-matas/
prashp 5y ago
I love Avatar, but the analogies in that notebook were quite difficult to follow!
infogulch 5y ago
That's pretty neat and looks nice. I wonder how it compares to fastai's nbdev/nb2md? https://www.fast.ai/2020/01/20/nb2md/ https://www.fast.ai/2020/01/20/nb2md/
- the21st 5y ago
  Hi, I'm Simon (Engineer at Deepnote). Thanks for the feedback! nbdev is actually orthogonal to what we're doing, and you can use it within Deepnote! I've written a short guide on it: https://deepnote.com/@the21st/nbdev-Deepnote-MQQVcO8aQV--6rCKUm-ftg https://deepnote.com/@the21st/nbdev-Deepnote-MQQVcO8aQV--6rC...
- jph00 5y ago
  Thanks for mentioning nbdev (which as mentioned works well with DeepNote). FYI, the blog post you linked to is a bit out of date - we have something much better for blogging with jupyter notebooks nowadays, which is fastpages: https://fastpages.fast.ai/ https://fastpages.fast.ai/ . It's compatible with the same annotations used in nbdev.
gk1 5y ago
Nice gallery and great idea to publish these with a "Launch in Deepnote" button. How is this different from publishing notebooks on Google Colab? Also, I'm seeing some rendering issues where my progress bar is printed repeatedly when it's supposed to get overwritten: https://deepnote.com/viewer/github/pinecone-io/examples/blob/master/image_search/simple_pytorch_image_search.ipynb https://deepnote.com/viewer/github/pinecone-io/examples/blob... Here's the Google Colab version for comparison: https://colab.research.google.com/github/pinecone-io/examples/blob/master/image_search/simple_pytorch_image_search.ipynb https://colab.research.google.com/github/pinecone-io/example... Edit: Looks like it's an issue inside the notebook, not with Deepnote. Nevermind!
- the21st 5y ago
  Thanks for the feedback! The main difference between us (Deepnote) and Colab is that we aim for a clean reading experience, and we optimize for speed. Colab seems to load a notebook that has editable cells and you can start executing cells directly, so it doesn't feel as a "publishing" feature, more like straight up notebook sharing. And thanks for the bug report, will look into that
  - adlha 5y ago
    Thanks for the q! A follow-up comment here - if you look away from the publishing feature, the experience of Deepnote vs. Colab mainly differs in a) UI, b) breadth of integrations (Deepnote integrates with most of the data sources out there and plays well with the rest of your stack) and c) unlike Colab, Deepnote supports both real-time collaboration and asynchronous collaboration via comments - so quite literally Google docs / Figma meets notebooks.
bachmeier 5y ago
One thing I wished they wouldn't do: > Deepnote is completely free. That's simply not true. The trial plan is completely free. Same as a gazillion other services. By all means, they should charge for their services and get rich, but Deepnote is not completely free.
- adlha 5y ago
  Hey, Liz from the Deepnote team here. We don't actually have a trial plan. If you're an individual or a small team, you can use our standard plan for free forever. You only upgrade if you want to support a larger team, get stronger hardware or have Enterprise requirements. I'm sending more info about our plans & pricing here: https://deepnote.com/pricing https://deepnote.com/pricing
  - virgilp 5y ago
    The objections here would be trivially resolved if the pricing page said "Deepnote Standard is and always will be free" instead of "Deepnote is and always will be free".
    - benatkin 5y ago
      Well, they aren't the first to say "always free" despite having features that can be paid for, and once you get used to this little piece of marketing speak, it's easy to understand. I appreciate them using something commonly understood.
- KMnO4 5y ago
  The go on to say, > Upgrade if you need advanced workflows or fine-grained permission control. Okay, so I need only upgrade for those features? Except wait, you only get 3 projects in the free tier. Three! I would be more okay calling it a free service if 99% of users wouldn't need the paid tiers. But that tier is definitely just a trial.
  - adlha 5y ago
spicyramen 5y ago
This looks awesome, how does Deep note compares to Enterprise notebook solutions like Google Notebooks, Sagemaker or Databricks. My company cares about data exfiltration, PII data, CMEK. Our researchers have access to data that is very valuable and posting a Notebook that may render some information could be a problem, is there a way to render notebooks (stored in a private GitHub) to only a subset of authenticated users?
- robertlacok 5y ago
  Hey there! Product Manager from Deepnote here. We're in a stage where we support a few enterprises already, but so far only those who are comfortable using a managed service. However, the support for on-prem (deploying to your cloud environment) is coming soon and that should cover the majority of use cases. Just to make it a bit clearer – this is a fun tool to render any notebooks on GitHub. Deepnote as a platform also supports everything else you normally do with notebooks (execution, collaborative editing & comments, versioning, secure integrations, scheduling and more). And yes, you can set sharing settings similar to Google Docs, for each project (which can have one or more notebooks). CMEK are not yet available and not yet in the plan for v1 of on-prem, but let's talk (robert at deepnote.com), and we can make it happen.
matsemann 5y ago
I've been using a notebook running on Deepnote[0] to follow along to Andrew Ng's Coursera course (as I wanted to do it in python and not octave/matlab). It has worked really nice. [0]: https://github.com/dibgerge/ml-coursera-python-assignments https://github.com/dibgerge/ml-coursera-python-assignments
MartinMond 5y ago
Reminds me of https://github.nextjournal.com/ https://github.nextjournal.com/ where you can just load a url from GitHub: https://github.nextjournal.com/uwdata/visualization-curriculum/blob/master/altair_cartographic.ipynb https://github.nextjournal.com/uwdata/visualization-curricul...
reslan 5y ago
I find Deepnote very useful for showing my work to others (especially non-tech) people. This would really improve the way we share our findings, which so far, wasn't easy. I built a notebook on using different exploratory data analysis techniques here: https://deepnote.com/@reslan-al-tinawi/Visualizing-data-with-seaborn-plotly-2hf4mb-sTnC9LyUXbTVKDw https://deepnote.com/@reslan-al-tinawi/Visualizing-data-with...

217 points by adlha 5y ago

Hey everyone, our team just built a tool to make your Github notebooks interactive and we would love for you to try it out. Just enter the GitHub URL of your (or any) public notebook and hit Render to instantly view the notebook. Your notebook will be rendered in an article-like layout and will get a table of contents. Anyone can fork it using the "Launch" button and play around with your code. The Deepnote viewer is also faster and more reliable than other .ipynb viewers we've tried. Would love for you to take it for a spin & hear your feedback!

NmAmDa 5y ago
That seems great, but for the comparison between DN viewer and the other, is there any benchmark tests specifically or it just a general feeling of speed when someone is using them. Because for each Jupyter notebook viewer it will depend on the setup, machine and the load itself.
- the21st 5y ago
  Hi there! Simon (Engineer at Deepnote) here. We didn't benchmark the load speed but for some reason, Github's ipynb viewer has always felt to me quite slow and unreliable. All viewers are publicly accessible so I'd love it someone did an independent benchmark. I'd prefer to avoid doing one ourselves because of the obvious conflict of interest
  - verdverm 5y ago
    Why not make a benchmark which is also open and reproducible?
  - NmAmDa 5y ago
    I thought that it is more than this. I think you should support any claim about that by creating a reproducible open benchmark test so at least people can understand what criteria you selected.
    - NmAmDa 5y ago
      I also find it myself much faster than GitHub viewer.
erikreinertsen 5y ago
I am a huge fan of Deepnote! Have you considered / is there already a way to “import” an .ipynb into a DeepNote project? Would be easier for users than to copy and paste code one cell at a time.
- adlha 5y ago
  Absolutely! All you have to do is Upload your .ipynb files (or full folders) into an empty Deepnote project and run the code. I'm sending docs with more info on how to import files here: https://docs.deepnote.com/importing-and-exporing/importing-data-to-deepnote https://docs.deepnote.com/importing-and-exporing/importing-d...
carreau 5y ago
What's your security model for logged-in users, one of the reason we (the jupyter team) have nbviewer on a separate domain with no login is to have embedded JS and other potentially sensitive content to be rendered without risk. We've seen people trying many attack vectors against renderer like this one with for example injecting script tags in things like prompt numbers.
- sunaden 5y ago
  Thanks for the q (I work at Deepnote) - all outputs that can contain potentially malicious JS are sandboxed in iframes so they can only access their local context and can't be used e.g. for XSS attacks.

codeknight11 5y ago

I've been using Deepnote for a while and highly recommend it. I love the features and they have also built a great community. Github's jupyter rendering doesn't work for me most of the time so I am using Deepnote viewer these days. Check out my published article that I wrote using Deepnote a while ago - https://deepnote.com/@tj/The-Starry-Cat-7HRaWnp8QhWYeK_x_QLD9Q https://deepnote.com/@tj/The-Starry-Cat-7HRaWnp8QhWYeK_x_QLD...

dmytton 5y ago

I've been using Deepnote to do some energy modelling for an academic paper I'm writing, and it's been great. The UI is much nicer than any other Jupyter editor I've used, and the integration with GitHub works well. Just started collaborating with some other researchers on another project so the real-time editing is coming in handy. So much better than using Excel. I also interviewed their CTO, Jan Matas, as part of my developer interview services about how it's all built around k8s behind the scenes: https://console.dev/interviews/deepnote-jan-matas/ https://console.dev/interviews/deepnote-jan-matas/

prashp 5y ago

I love Avatar, but the analogies in that notebook were quite difficult to follow!

infogulch 5y ago

That's pretty neat and looks nice. I wonder how it compares to fastai's nbdev/nb2md? https://www.fast.ai/2020/01/20/nb2md/ https://www.fast.ai/2020/01/20/nb2md/

the21st 5y ago
Hi, I'm Simon (Engineer at Deepnote). Thanks for the feedback! nbdev is actually orthogonal to what we're doing, and you can use it within Deepnote! I've written a short guide on it: https://deepnote.com/@the21st/nbdev-Deepnote-MQQVcO8aQV--6rCKUm-ftg https://deepnote.com/@the21st/nbdev-Deepnote-MQQVcO8aQV--6rC...
jph00 5y ago
Thanks for mentioning nbdev (which as mentioned works well with DeepNote). FYI, the blog post you linked to is a bit out of date - we have something much better for blogging with jupyter notebooks nowadays, which is fastpages: https://fastpages.fast.ai/ https://fastpages.fast.ai/ . It's compatible with the same annotations used in nbdev.

gk1 5y ago

Nice gallery and great idea to publish these with a "Launch in Deepnote" button. How is this different from publishing notebooks on Google Colab? Also, I'm seeing some rendering issues where my progress bar is printed repeatedly when it's supposed to get overwritten: https://deepnote.com/viewer/github/pinecone-io/examples/blob/master/image_search/simple_pytorch_image_search.ipynb https://deepnote.com/viewer/github/pinecone-io/examples/blob... Here's the Google Colab version for comparison: https://colab.research.google.com/github/pinecone-io/examples/blob/master/image_search/simple_pytorch_image_search.ipynb https://colab.research.google.com/github/pinecone-io/example... Edit: Looks like it's an issue inside the notebook, not with Deepnote. Nevermind!

the21st 5y ago
Thanks for the feedback! The main difference between us (Deepnote) and Colab is that we aim for a clean reading experience, and we optimize for speed. Colab seems to load a notebook that has editable cells and you can start executing cells directly, so it doesn't feel as a "publishing" feature, more like straight up notebook sharing. And thanks for the bug report, will look into that
- adlha 5y ago
  Thanks for the q! A follow-up comment here - if you look away from the publishing feature, the experience of Deepnote vs. Colab mainly differs in a) UI, b) breadth of integrations (Deepnote integrates with most of the data sources out there and plays well with the rest of your stack) and c) unlike Colab, Deepnote supports both real-time collaboration and asynchronous collaboration via comments - so quite literally Google docs / Figma meets notebooks.

bachmeier 5y ago

One thing I wished they wouldn't do: > Deepnote is completely free. That's simply not true. The trial plan is completely free. Same as a gazillion other services. By all means, they should charge for their services and get rich, but Deepnote is not completely free.

adlha 5y ago
Hey, Liz from the Deepnote team here. We don't actually have a trial plan. If you're an individual or a small team, you can use our standard plan for free forever. You only upgrade if you want to support a larger team, get stronger hardware or have Enterprise requirements. I'm sending more info about our plans & pricing here: https://deepnote.com/pricing https://deepnote.com/pricing
- virgilp 5y ago
  The objections here would be trivially resolved if the pricing page said "Deepnote Standard is and always will be free" instead of "Deepnote is and always will be free".
  - benatkin 5y ago
    Well, they aren't the first to say "always free" despite having features that can be paid for, and once you get used to this little piece of marketing speak, it's easy to understand. I appreciate them using something commonly understood.
KMnO4 5y ago
The go on to say, > Upgrade if you need advanced workflows or fine-grained permission control. Okay, so I need only upgrade for those features? Except wait, you only get 3 projects in the free tier. Three! I would be more okay calling it a free service if 99% of users wouldn't need the paid tiers. But that tier is definitely just a trial.
- adlha 5y ago

spicyramen 5y ago

This looks awesome, how does Deep note compares to Enterprise notebook solutions like Google Notebooks, Sagemaker or Databricks. My company cares about data exfiltration, PII data, CMEK. Our researchers have access to data that is very valuable and posting a Notebook that may render some information could be a problem, is there a way to render notebooks (stored in a private GitHub) to only a subset of authenticated users?

robertlacok 5y ago
Hey there! Product Manager from Deepnote here. We're in a stage where we support a few enterprises already, but so far only those who are comfortable using a managed service. However, the support for on-prem (deploying to your cloud environment) is coming soon and that should cover the majority of use cases. Just to make it a bit clearer – this is a fun tool to render any notebooks on GitHub. Deepnote as a platform also supports everything else you normally do with notebooks (execution, collaborative editing & comments, versioning, secure integrations, scheduling and more). And yes, you can set sharing settings similar to Google Docs, for each project (which can have one or more notebooks). CMEK are not yet available and not yet in the plan for v1 of on-prem, but let's talk (robert at deepnote.com), and we can make it happen.

matsemann 5y ago

I've been using a notebook running on Deepnote[0] to follow along to Andrew Ng's Coursera course (as I wanted to do it in python and not octave/matlab). It has worked really nice. [0]: https://github.com/dibgerge/ml-coursera-python-assignments https://github.com/dibgerge/ml-coursera-python-assignments

MartinMond 5y ago

Reminds me of https://github.nextjournal.com/ https://github.nextjournal.com/ where you can just load a url from GitHub: https://github.nextjournal.com/uwdata/visualization-curriculum/blob/master/altair_cartographic.ipynb https://github.nextjournal.com/uwdata/visualization-curricul...

reslan 5y ago

I find Deepnote very useful for showing my work to others (especially non-tech) people. This would really improve the way we share our findings, which so far, wasn't easy. I built a notebook on using different exploratory data analysis techniques here: https://deepnote.com/@reslan-al-tinawi/Visualizing-data-with-seaborn-plotly-2hf4mb-sTnC9LyUXbTVKDw https://deepnote.com/@reslan-al-tinawi/Visualizing-data-with...