fyi, I updated the boolean indexing to support additional or missing dimensions: https://gist.github.com/Hoeze/96616ef9d179180b0b7de97c97e00a27 I'm using this on a 4D-array with >300GB to flatten three of the four dimensions and it works, even on 64GB of RAM.

@max-sixty and I have been having some more discussion about whether this is what ds[key] should do for N-dimensional boolean indexing over in #1887.

But regardless of what we want boolean indexing with [] to do, this would certainly be welcome functionality and should exist in a dedicated method. ds[key] is already very heavily overloaded in Xarray, so a more explicit option is nice to have, e.g., for the benefit of readability and static type checking. For the same reason, I would rather not put it inside isel() which already integer based indexing with a different call signature. My tentative suggestion is to call this new method sel_mask(), since that's what it does -- selection like sel/isel except based on a mask.

I wonder if this is just a better proposal than making N-dimensional boolean indexing an alias for where: https://github.com/pydata/xarray/issues/1887#issuecomment-823673654

Ah right, I see now, thanks for explaining.

Allowing pointwise indexing with bool indexes would also be welcome.

@max-sixty The reason is that my method is basically a special case of point-wise indexing: http://xarray.pydata.org/en/stable/indexing.html#more-advanced-indexing You can get the same result by calling: ```python core_dim_locs = {key: value for key, value in core_dim_locs_from_cond(mask, new_dim_name="newdim")}

pointwise selection

data.sel( dim_0=outliers_subset["dim_0"], dim_1=outliers_subset["dim_1"], dim_2=outliers_subset["dim_2"] ) ``` (Note that you loose chunk information by this method, that's why it is less efficient)

When you want to select random items from a N-dimensional array, you can either model the result as some sparse array or by stacking the dimensions. (OK, stacking the dimensions means also a sparse COO encoding...)

Thanks for the issue @Hoeze . Multi-dimensional bool indexing is definitely something we'd like to add.

How does your code differ from the proposals in https://github.com/pydata/xarray/issues/1887? In a brief look through the code — thanks for supplying it — I couldn't immediately see why we need a new dimension?