issue_comments

Also, just for future reference usually we close feature requests when the corresponding PR has been merged :)

On Fri, Apr 1, 2016 at 11:04 AM, Phillip Wolfram notifications@github.com wrote:

Closed #811 https://github.com/pydata/xarray/issues/811.

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub https://github.com/pydata/xarray/issues/811#event-611609243

If I'm understanding you correctly, the use case is to trim off all NA regions when using where for selection. If your mask is not perfectly rectangular, some values will still be replaced with NA.

A starting point would be something like this:

python def sel_where(data, mask): data, mask = xr.broadcast(xr.align(data, mask, join='left', copy=False)) # possibly not a good idea to expand mask to the full dimensions of the data indexes = tuple(mask.any(dim).values for dim in mask.dims) return data[indexes].where(mask[indexes])

I'm still not sure exactly what you mean by this "contraction" like behavior. Could you write this out in pseudo code?

This is pretty different from .sel and not in the way that reindex_like is different from reindex, so possibly another name would be appropriate.

I would rather make a sel_like method for remapping acase.sel_like(idx) -> acase.sel(x=idx.x.values, y=idx.y.values). That would mirror the current reindex/reindex_like distinction, though reindex_like does solve many of these use cases.

If acase and idx (a boolean) both have dimensions ('x', 'y'), what should acase.sel(idx) or acase[idx] do?

NumPy will return a 1D flattened array in this case. We could do that (with a MultiIndex) but that's not so useful in xarray.

You can write acase.where(idx), but we don't support acase.sel(idx) because it's not clear what do to if idx has more than one dimension.

I suppose we could allow acase.sel(idx1, idx2, ...) if all the provided arguments are 1D.

Should xarray indexing account for boolean values without resorting to a call to np.where?

as far as I know, it does:

python In [1]: import xarray as xr In [2]: import numpy as np In [3]: da = xr.DataArray(np.arange(10), coords={'time':np.arange(10)}) In [4]: da.sel(time=da.time > 4) Out[4]: <xarray.DataArray (time: 5)> array([5, 6, 7, 8, 9]) Coordinates: * time (time) int64 5 6 7 8 9

But according to the traceback it seems to have something to do with the shape or your array?