3 ms·

My gripe with Observable is I can't fit into a Python-based workflow. I.e., a machine learning or gpu-training worfklow. It is a nice place to create explainers

My gripe with Observable is I can't fit into a Python-based workflow. I.e., a machine learning or gpu-training worfklow. It is a nice place to create explainers, though. I almost think of it as being like a Medium alternative.

fractionalhare 6y ago
This is also the main thing holding me back from using it. The Observable experience feels more polished than Jupyter for me.
- petulla 6y ago
  My hope is eventually to be able to run it locally and integrate it into other language, like Tensorboard.
catacombs 6y ago
> My gripe with Observable is I can't fit into a Python-based workflow. Observable is for JavaScript ecosystem. I wouldn't count on them implementing anything for Python.
- hobofan 6y ago
  With the growth expectations that come with the funding amount they just collected, I'd count on it. There is just so much more demand for data visualization and notebook-style projects in the Python community compared to the JS community.
pea 6y ago
For Python, Datapane may solve your use-case (I am one of the founders). It's an open source framework for creating HTML reports from text, dataframes, plots (altair/bokeh/plotly/folium), and files. We provide a free public platform for sharing reports on the web, and you can check out some of the things our users are building here: https://datapane.com/gallery/ https://datapane.com/gallery/ It's different to something like Observable in that you import it into your existing Python environment (Colab, Jupyter, Airflow, etc.), build the reports from your assets, and push them up -- as opposed to having to do the analysis itself in our interface.
domoritz 6y ago
Take a look at https://www.streamlit.io/ https://www.streamlit.io/.
- petulla 6y ago
  hey Dominik Yes I'm a fan; just would like to be able to write JS to visualize.
ilaksh 6y ago
Can you run arbitrary libraries like Tensorflow.js?
- TeMPOraL 6y ago
  You can run just about anything that can be run in the browser (all code in an Observable notebook executes in your browser). Though depending on the flavour of packaging/module system a given package uses, you may need to figure out a correct magic incantation to import it.
Jugurtha 6y ago
Could you say more about your workflow? Our platform[0] may suit you: - Collaborative no-setup notebooks: different images with the most popular libraries pre-installed. See each other's cursors, very useful for troubleshooting, pair programming. - Long-running notebooks: GPU ready. You can schedule notebooks right from the notebook view so you don't context switch. You can close your browser, disconnect from the internet, or shut your laptop and still see the notebook's execution. - Automatic params/metrics/model tracking: no boilerplate. - One click deployment: clicking a button deploys a model and gives you a nice "REST" API endpoint you can invoke. If you're doing anything related to ML and use notebooks, you probably had the above problems. We're building around Jupyter notebooks because in our experience, our machine learning projects were not held back for lack of slick stylesheets. - [0]: https://iko.ai https://iko.ai

fractionalhare 6y ago

This is also the main thing holding me back from using it. The Observable experience feels more polished than Jupyter for me.

petulla 6y ago
My hope is eventually to be able to run it locally and integrate it into other language, like Tensorboard.

catacombs 6y ago

> My gripe with Observable is I can't fit into a Python-based workflow. Observable is for JavaScript ecosystem. I wouldn't count on them implementing anything for Python.

hobofan 6y ago
With the growth expectations that come with the funding amount they just collected, I'd count on it. There is just so much more demand for data visualization and notebook-style projects in the Python community compared to the JS community.

pea 6y ago

For Python, Datapane may solve your use-case (I am one of the founders). It's an open source framework for creating HTML reports from text, dataframes, plots (altair/bokeh/plotly/folium), and files. We provide a free public platform for sharing reports on the web, and you can check out some of the things our users are building here: https://datapane.com/gallery/ https://datapane.com/gallery/ It's different to something like Observable in that you import it into your existing Python environment (Colab, Jupyter, Airflow, etc.), build the reports from your assets, and push them up -- as opposed to having to do the analysis itself in our interface.

domoritz 6y ago

Take a look at https://www.streamlit.io/ https://www.streamlit.io/.

petulla 6y ago
hey Dominik Yes I'm a fan; just would like to be able to write JS to visualize.

ilaksh 6y ago

Can you run arbitrary libraries like Tensorflow.js?

TeMPOraL 6y ago
You can run just about anything that can be run in the browser (all code in an Observable notebook executes in your browser). Though depending on the flavour of packaging/module system a given package uses, you may need to figure out a correct magic incantation to import it.

Jugurtha 6y ago

Could you say more about your workflow? Our platform[0] may suit you: - Collaborative no-setup notebooks: different images with the most popular libraries pre-installed. See each other's cursors, very useful for troubleshooting, pair programming. - Long-running notebooks: GPU ready. You can schedule notebooks right from the notebook view so you don't context switch. You can close your browser, disconnect from the internet, or shut your laptop and still see the notebook's execution. - Automatic params/metrics/model tracking: no boilerplate. - One click deployment: clicking a button deploys a model and gives you a nice "REST" API endpoint you can invoke. If you're doing anything related to ML and use notebooks, you probably had the above problems. We're building around Jupyter notebooks because in our experience, our machine learning projects were not held back for lack of slick stylesheets. - [0]: https://iko.ai https://iko.ai