I would appreciate if you clarified what Dask is at the beginning of the post. I'm a Python developer and it's the first time I hear about it and I had to scroll pretty much to the end for a link to its documentation.
The target audience isn't all Python developers, but specifically data scientists or engineers who work with the PyData stack, to whom Dask is pretty well known.
Has anyone here had positive experiences with Dask?
I chose it as an alternative to manual batching, where a dataset couldn't fit into memory. Anecdotally, it was great when it worked, but caused a lot of random bugs, and I spent more time fixing them than the time it saved in the first place. (This was mid-2019 so maybe it has improved since then)
I generally share your criticism, but my overall experience has been positive. Dask indeed requires some fiddling, and it's API and documentation has some blind spots, but I've successfully used it to scale up some Pandas code to run on dataframes measuring in the tens of GB. Being able to transfer my knowledge of Pandas and NumPy so readily makes Dask's speed bumps worth it.
It would be more true to GVR's original inspiration for the language if Python-based businesses named themselves things like "Ministry of Silly Walks", "The Antlers of Dictation", "Tobacconist", or "Bevis, Be Gentle!" rather than "Coiled", "Anaconda", etc. ;-)
11 comments
[ 3.0 ms ] story [ 34.7 ms ] thread> Dask provides advanced parallelism for analytics, enabling performance at scale for the tools you love
edit: removed comments
https://yourdatafitsinram.net/
I chose it as an alternative to manual batching, where a dataset couldn't fit into memory. Anecdotally, it was great when it worked, but caused a lot of random bugs, and I spent more time fixing them than the time it saved in the first place. (This was mid-2019 so maybe it has improved since then)