> Dask provides advanced parallelism for analytics, enabling performance at scale for the tools you love
edit: removed comments
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)