Hi this is great. Having the notebook in Snowflake. However, I tried to run a simple fget_object from minio package, and got an error. Same code that runs fine in jupyter notebook.
I went through the video and I can say with confidence that the solution is not built with the data science methodology in mind (like all similar efforts by companies like snowflake). In fact, it was probably designed by people with very limited experience on the DS product lifecycle. To name a few of my objections: 1. Non-existent dependency management. Just force-fed the Anaconda packages and versions Snowflake likes. 2. By moving away from Jupyter, we also lose all the myriads of cool extensions, including the ability to write and export reports to pdfs. 3. The unclear combination of SQL, SnowPark and Python/Pandas, will make the productionisation of any notebook impossible. The whole solution will have to be rewritten from scratch. 4. No ability for external files or a dir structure. Good luck building a POC, without having several custom modules loaded into the notebook. 5. Adding streamlit into the mix is simply incomprehensible. How are we gonna enforce the reproducibility of our reports if people will just interact with the streamlit components? Snowflake reintroduces the same problems that using Excel has. The lack of a cohesive and structured sequence of transformations. 6. I generally don't understand the need to simplify and convolute the normal engineering process. You can't be a data scientist in 2024 if you don't know Python, why do you keep creating such a mess? All in all, who is the target audience and the target use case? Who in the right mind will run production systems from a notebook? And you did all of the above without tackling the actual 3 most important limitations of jupyter notebooks: A. Collaborating and commenting B. The ability to do effective peer reviews C. The lack of linting, formatting and autocompleting, so that the generated code adheres to the Python community standards. I am sorry, but that's a disappointing release, similar to the implementation of streamlit apps. IF you want to build good tools that become the new norm, you need to hire data scientists from the industry who know what the actual day-to-day needs of a serious data science team are.
Thank you so much for the detailed feedback! We're actively working on making Snowflake Notebooks a best-in-class experience, so definitely keep an eye out for future improvements.
Cool feature but you are basically recreating databricks notebooks which are waaaaaay more advanced. they already have an llm integrated and you dont have to define an ugly df = cell5 command
Great videos! Hopefully this of videos on Notebooks means they're close to PuPr :-)
Really looking forward to seeing what we can do with them!
Hi this is great. Having the notebook in Snowflake. However, I tried to run a simple fget_object from minio package, and got an error. Same code that runs fine in jupyter notebook.
I went through the video and I can say with confidence that the solution is not built with the data science methodology in mind (like all similar efforts by companies like snowflake). In fact, it was probably designed by people with very limited experience on the DS product lifecycle.
To name a few of my objections:
1. Non-existent dependency management. Just force-fed the Anaconda packages and versions Snowflake likes.
2. By moving away from Jupyter, we also lose all the myriads of cool extensions, including the ability to write and export reports to pdfs.
3. The unclear combination of SQL, SnowPark and Python/Pandas, will make the productionisation of any notebook impossible. The whole solution will have to be rewritten from scratch.
4. No ability for external files or a dir structure. Good luck building a POC, without having several custom modules loaded into the notebook.
5. Adding streamlit into the mix is simply incomprehensible. How are we gonna enforce the reproducibility of our reports if people will just interact with the streamlit components? Snowflake reintroduces the same problems that using Excel has. The lack of a cohesive and structured sequence of transformations.
6. I generally don't understand the need to simplify and convolute the normal engineering process. You can't be a data scientist in 2024 if you don't know Python, why do you keep creating such a mess?
All in all, who is the target audience and the target use case?
Who in the right mind will run production systems from a notebook?
And you did all of the above without tackling the actual 3 most important limitations of jupyter notebooks:
A. Collaborating and commenting
B. The ability to do effective peer reviews
C. The lack of linting, formatting and autocompleting, so that the generated code adheres to the Python community standards.
I am sorry, but that's a disappointing release, similar to the implementation of streamlit apps. IF you want to build good tools that become the new norm, you need to hire data scientists from the industry who know what the actual day-to-day needs of a serious data science team are.
Thank you so much for the detailed feedback! We're actively working on making Snowflake Notebooks a best-in-class experience, so definitely keep an eye out for future improvements.
Cool feature but you are basically recreating databricks notebooks which are waaaaaay more advanced. they already have an llm integrated and you dont have to define an ugly df = cell5 command