Sparse DataArray Xarray search

Question

Using DataArray objects in xarray what is the best way to find all cells that have values != 0.

For example in pandas I would do

df.loc[df.col1 > 0]

My specific example I'm trying to look at 3 dimensional brain imaging data.

first_image_xarray.shape
(140, 140, 96)
dims = ['x','y','z']

Looking at the documentation for xarray.DataArray.where it seems I want something like this:

first_image_xarray.where(first_image_xarray.y + first_image_xarray.x  > 0,drop = True)[:,0,0]

But I still get arrays with zeros.

<xarray.DataArray (x: 140)>
array([ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0., -0.,  0., -0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.])
Dimensions without coordinates: x

Also - a side question - why are there some negative zeros? Are these values rounded and -0. is actually equal to something like -0.009876 or something?

Answer 1

(Answer to main question)

You are almost there. However, a slight syntax difference makes a big difference here. On one hand, here is the solution to filter >0 values using a "value-based" mask.

# if you want to DROP values which do not suffice a mask condition
first_image_xarray[:,0,0].where(first_image_xarray[:,0,0] > 0, drop=True)

or

# if you want to KEEP values which do not suffice a mask condition as nan
first_image_xarray[:,0,0].where(first_image_xarray[:,0,0] > 0, np.nan)

On the other hand, the reason why your attempt did not work as you hoped is because with first_image_xarray.x, it is referring to the index of elements in the array (in x direction) rather than referring to the value of the elements. Thus only the 1st element of your output should be nan instead of 0 because it only does not suffice the mask condition in slice [:,0,0]. Yes, you were creating an "index-based" mask.

The following small experiment (hopefully) articulates this critical difference.

Suppose we have DataArray which consists of only 0 and 1 (dimension is aligned with the original post (OP) of the question (140,140,96)). Firstly let's mask it based on index as OP did:

import numpy as np
import xarray as xr

np.random.seed(0)
# create a DataArray which randomly contains 0 or 1 values
a = xr.DataArray(np.random.randint(0, 2, 140*140*96).reshape((140, 140, 96)), dims=('x', 'y', 'z'))


# with this "index-based" mask, only elements where index of both x and y are 0 are replaced by nan
a.where(a.x + a.y > 0, drop=True)[:,0,0]

Out:
<xarray.DataArray (x: 140)>
array([ nan,   0.,   1.,   1.,   0.,   0.,   0.,   1.,   0.,   0.,   0.,   0.,
         0.,   1.,   0.,   1.,   0.,   1.,   0.,   0.,   0.,   1.,   0.,   0.,
         1.,   1.,   0.,   1.,   1.,   1.,   1.,   1.,   1.,   1.,   1.,   1.,
         1.,   1.,   0.,   1.,   1.,   1.,   1.,   1.,   1.,   1.,   0.,   1.,
         1.,   0.,   0.,   0.,   1.,   1.,   1.,   0.,   0.,   1.,   0.,   0.,
         1.,   0.,   1.,   1.,   0.,   0.,   1.,   0.,   0.,   1.,   1.,   1.,
         0.,   0.,   0.,   1.,   1.,   0.,   1.,   0.,   1.,   1.,   0.,   0.,
         0.,   0.,   1.,   1.,   0.,   1.,   1.,   1.,   1.,   0.,   1.,   0.,
         0.,   0.,   0.,   0.,   0.,   0.,   1.,   0.,   1.,   1.,   0.,   0.,
         0.,   0.,   1.,   0.,   1.,   0.,   0.,   0.,   0.,   1.,   0.,   1.,
         0.,   0.,   1.,   0.,   0.,   0.,   0.,   0.,   1.,   1.,   0.,   0.,
         0.,   1.,   0.,   0.,   1.,   0.,   0.,   1.])
Dimensions without coordinates: x

With the mask above, only the element where index of both x and y are 0 turns in to nan and the rest has not been changed or dropped at all.

In contrast, the proposed solution masks the DataArray based on the values of DataArray elements.

# with this "value-based" mask, all the values which do not suffice the mask condition are dropped
a[:,0,0].where(a[:,0,0] > 0, drop=True)

Out:
<xarray.DataArray (x: 65)>
array([ 1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,
        1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,
        1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,
        1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,
        1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.])
Dimensions without coordinates: x

This successfully dropped all the values which do not suffice a mask condition based on the values of DataArray elements.

---

(Answer to side question)

As for the origin of -0 and 0 in DataArray, rounded values from negative or positive side towards 0 would be the possibility: A related discussion was done here How to eliminate the extra minus sign when rounding negative numbers towards zero in numpy? The below is a tiny example of this case.

import numpy as np
import xarray as xr

xr_array = xr.DataArray([-0.1, 0.1])

# you can use either xr.DataArray.round() or np.round() for rounding values of DataArray

xr.DataArray.round(xr_array)

Out:
<xarray.DataArray (dim_0: 2)>
array([-0.,  0.])
Dimensions without coordinates: dim_0

np.round(xr_array)

Out:
<xarray.DataArray (dim_0: 2)>
array([-0.,  0.])
Dimensions without coordinates: dim_0

As a side note, the other possibility for getting -0 in NumPy array can be numpy.set_printoptions(precision=0), which hides below decimal point like below (but I know this is not the case this time since you are using DataArray):

import numpy as np

# default value is precision=8 in ver1.15
np.set_printoptions(precision=0)

np.array([-0.1, 0.1])

Out:
array([-0.,  0.])

Anyway, My best guess is that the conversion to -0 should be manual and intentional rather than automatic in data preparation & pre-processing phase.

Hope this helps.