issue_comments

I wake up this issue, Any news ?

A possible code for this functionality could be: def nonzero4xarray (da:xarray.DataArray): c = da.where(da, drop=True).coords s = da.sel(c).to_series() s = s[s] o = [] for record in s.index: o.append(dict(list(zip(s.index.names, record)))) return o however the output is sub-optimal, as it yields a list of dicts, each with the corresponding coordinate to the nonzero value.

Consider the following example: ```

import numpy as np import xarray as xr x = np.arange(5) * 100 y = np.arange(4) * 100 z = np.arange(3) * 10 da = xr.DataArray(np.random.randn(z.size, y.size, x.size), dims=('z', 'y', 'x'), coords={'x': x, 'y': y, 'z': z}) da.loc[da.isel({'x':[0, 2], 'y':[1,3], 'z':[0]}).coords] = np.nan da.loc[da.isel({'x':[1, 3], 'y':[0], 'z':[-1]}).coords] = np.nan and imagine I want to capture the locations with NaN values. If I do: coords = da.where(da.isnull(), drop=True).coords da.sel(coords) array([[[ 1.586261, -0.2003 , 0.734872, -1.198746], [ nan, -0.108265, nan, 0.754857], [ nan, 0.700742, nan, -1.514642]],

   [[ 1.109573,       nan, -1.596756,       nan],
    [ 0.031771, -0.341263,  1.358767,  0.221174],
    [-0.671605,  0.923703,  0.085335, -1.672686]]])

``` I will get the coordinates for a mix of NaN and valid values, depending on how these are distributed.

If I want to obtain only the NaN locations: ```

coords = nonzero4xarray(da.isnull()) coords [{'z': 0, 'y': 100, 'x': 0}, {'z': 0, 'y': 100, 'x': 200}, {'z': 0, 'y': 300, 'x': 0}, {'z': 0, 'y': 300, 'x': 200}, {'z': 20, 'y': 0, 'x': 100}, {'z': 20, 'y': 0, 'x': 300}] vals = [da.sel(c).values[()] for c in coords] vals [nan, nan, nan, nan, nan, nan] `` whose functionality is similar tonumpy.nonzero, although sub-optimal as we cannot issueda.sel(coords)` directly.

In order to maintain a list of currently relevant issues, we mark issues as stale after a period of inactivity

If this issue remains relevant, please comment here or remove the stale label; otherwise it will be marked as closed automatically

For the above example, I consider the following output,

python <xarray.Dataset> Dimensions: (nz_points: 11) Dimensions without coordinates: nz_points Data variables: x (nz_points) int64 0 0 1 1 1 2 2 2 3 3 3 y (nz_points) int64 1 2 0 1 2 0 1 2 0 1 2

It would return a xr.Dataset each keys of which are the dimension names and their values are xr.DataArray of valid indices. da[da.nonzero()] or da.isel(**da.nonzero()) would return a 1d-DataArray with non-dimensional coordinates, x and y. (nz_points is the default dimension, equivalent to da.nonzero(dim='nz_points'))

nonzero method for Dataset is not obvious, but the return of xr.DataArray.nonzero should be passed to xr.Dataset.isel().

What exactly should the result of DataArray.nonzero() or Dataset.nonzero() be?

In NumPy, nonzero() returns a tuple of 1D integer arrays suitable for indexing. It's not obvious to me how that generalizes to xarray.