Clip function with minimum or maximum nan value

[babameme]

Hi team,

1. Concise description about the problem: Clip function with minimum or maximum nan value

I see that with 1-dim DataArray, we can clip with minimum or maximum nan value:

xx = xr.DataArray([1, 3, 5], dims = "x")
xx.clip(np.nan, 4)

Result:

xarray.DataArrayx: 3
array([1., 3., 4.])
Coordinates: (0)
Attributes: (0)

But with 2-dims DataArray, if one of two bounds is np.nan, we will receive nan value:

da = xr.DataArray(np.arange(18).reshape(3, 6), coords = {"x": range(3), "y":range(6)}, dims = ("x", "y"))
lower = xr.DataArray([1, 2, np.nan], coords = {"x": [0, 1, 2]}, dims = ["x"])
upper = xr.DataArray([4, 10, 14], coords = {"x": [0, 1, 2]}, dims = ["x"])
lower = xr.broadcast(da, lower)[1]
upper = xr.broadcast(da, upper)[1]
da.clip(lower, upper)

Result:

xarray.DataArrayx: 3y: 6
array([[ 1., 1., 2., 3., 4., 4.],
[ 6., 7., 8., 9., 10., 10.],
[nan, nan, nan, nan, nan, nan]])
Coordinates:
x
(x)
int64
0 1 2
y
(y)
int64
0 1 2 3 4 5
Attributes: (0)

2. Describe the solution i would like:

The clip function perform well with 2-dims as it done in 1-dims

[max-sixty]

Hi @babameme — please could you fill out the issue template, including formatting the code reasonably?

[babameme]

Hi @max-sixty , this is the first time i request feature in github. Can you introduce how to fill out the issue template ?

[max-sixty]

Sure — when you created the issue, there was a template in the text box — could you fill that out?

[max-sixty]

If that's unclear, check out some other issues in the repo and use those as a template for yours.

[babameme]

Hi @max-sixty , i update my request
Let's check it,
Regards

[max-sixty]

It's still not great but it's sufficient!

What you want isn't available in numpy, the library we wrap:

In [3]: da.values.clip(lower.values, upper.values)
Out[3]:
array([[ 1.,  1.,  2.,  3.,  4.,  4.],
       [ 6.,  7.,  8.,  9., 10., 10.],
       [nan, nan, nan, nan, nan, nan]])

But I think what you want is this:

In [11]: lower_inf = lower.where(lower.notnull(), np.inf)

In [14]: lower_inf
Out[14]:
<xarray.DataArray (x: 3, y: 6)>
array([[ 1.,  1.,  1.,  1.,  1.,  1.],
       [ 2.,  2.,  2.,  2.,  2.,  2.],
       [inf, inf, inf, inf, inf, inf]])
Coordinates:
  * x        (x) int64 0 1 2
  * y        (y) int64 0 1 2 3 4 5

In [12]: da.clip(lower_inf, upper)
Out[12]:
<xarray.DataArray (x: 3, y: 6)>
array([[ 1.,  1.,  2.,  3.,  4.,  4.],
       [ 6.,  7.,  8.,  9., 10., 10.],
       [14., 14., 14., 14., 14., 14.]])
Coordinates:
  * x        (x) int64 0 1 2
  * y        (y) int64 0 1 2 3 4 5
  * ```

[babameme]

I got it. Thank you very much

[babameme]

Wait, do you see the result on the 3rd row:

[14., 14., 14., 14., 14., 14.]

The original row is

[12, 13, 14, 15, 16, 17]

the lower bound is inf, and the upper bound is 14.

I think the result should be:

[12, 13, 14, 14, 14, 14]

[max-sixty]

I got the sign wrong, corrected:

In [4]: lower_inf = lower.where(lower.notnull(), -np.inf)

In [5]: da.clip(lower_inf, upper)
Out[5]:
<xarray.DataArray (x: 3, y: 6)>
array([[ 1.,  1.,  2.,  3.,  4.,  4.],
       [ 6.,  7.,  8.,  9., 10., 10.],
       [12., 13., 14., 14., 14., 14.]])
Coordinates:
  * x        (x) int64 0 1 2
  * y        (y) int64 0 1 2 3 4 5