Julia Signell

I really like the idea of using collections this way. Historically have people proposed other approaches to this issue that were considered too high-level at the time?

Ping @eric-czech and @RogerMoens since you have both looked at the svd implementation recently.

If the error arises before you call `.compute` then the meta might be the issue, but I think what you are seeing is not an error, but incorrect results, so...

This looks a lot like the conversation around `__contains__` in https://github.com/dask/dask/pull/7914. I suspect that a _lot_ of code will break if we switch to returning a delayed object, so I...

Thanks for writing this up Tom! I agree with the two contradictory goals. > In general, my goal when I was doing pandas compatibility was to match latest version of...

It looks like the sort is only set up to apply across partitions in the aggregation step. This is a bug, but there aren't real performance implications when a dask...

ok to test

Just chiming in to say that I totally agree with the assessment of the current state of this code and I have a strong suspicion that even chunking just seemed...

It sounds like you have a good handle on the issue and how you think it should be solved. Probably the most productive next step would be to open a...

I wasn't aware of the `set_flags` options on pandas. Implementing the flag `allow_duplicate_labels` seems like a good solution that way people can opt in to. It seems like the general...