Noah D. Brenowitz

> So it works in theory, but because of the reshapes='feature' argument we lose all coordinates along the feature dimension. It should be pretty easy to pass along a pd.MultiIndex...

@darothen At first glance, that approach seems very similar to what I have been working on. Instead of `FieldExtractor`, `DatasetAdapter` , and `DatasetSelector`, I have `Select` and `Stacker` classes that...

Ok. I think I may understand now. I have always assumed the number of samples are constant between the input and output, but I can see how making the whole...

Cool! FWIW I have a couple transformers in [here](https://github.com/nbren12/sklearn-xarray/blob/master/sklearn_xarray/sklearn_xarray.py). At first glance it seems that @darothen and I differ on how we handle datasets vs dataarrays. I have been working...

> we'd need a better name though... How about `DataArrayUnion`? I do think the output of `DataArrayUnion.transform` should be a DataArray with a multiindex rather than a Dataset for the...

I definitely think the dictionary input idea is good, but I think it is better to provide it as a function sort of like how sklearn has `make_union` and `make_pipeline`....

It was fun meeting everyone today. I hope we can continue to talk about some of these issues together. I just profiled the send/recv performance of redis. Checkout https://gist.github.com/d7283b77634f52d6c1afd62c67bfc254. I...

I just added pyarrow. It's blazing fast as expected! [gist](https://gist.github.com/nbren12/598cb54aa60be2b105ced15149558bb8)

Stock data sounds like a good idea, although that might scare off some academic scientists, who tend to be a little snooty when it comes to finance. I am not...

That'd be pretty cool if we can access that feed!