operation on complex number data

[davidtrem]

Hello,

Would it be possible to add complex function operation on xray.DataArray ?

Function like np.sin, np.abs works properly on xray.DataArray containing complex numbers and
returns a DataArray (as expected).
However, np.real, np.imag, np.angle operation return an numpy.ndarray array instead of DataArray...

[davidtrem]

A short example for convenience:

import xray
import numpy as np
arr = xray.DataArray(np.random.randn(2, 3) + 1j*np.random.randn(2, 3))
np.abs(arr)  # returns a xray DataArray
np.angle(arr)  # returns a numpy ndarray 

[davidtrem]

I realized this is also an issue in Pandas. So I also reported here:
pandas-dev/pandas#10921

[shoyer]

There's nothing we can do about np.angle, np.real and np.imag unfortunately -- these need to be updated on the numpy side to call __array_wrap__.

However, we could certainly add .real and .imag properties that access these computed values as DataArrays.

[davidtrem]

That would be great to add .real and .imag .abs and .angle 👍

[rth]

There is an open issue at numpy about this in numpy/numpy#6266

Also, for future reference, locally re-defining np.angle by removing the z = array(z) line from the official function appears to work well enough as a workaround, assuming the input is an xarray,

import numpy.core.numeric as _nx


def angle(z, deg=0):
    """Compute the angle of an xarray

    Parameters
    ----------
    z : array_like
        A complex number or sequence of complex numbers.
    deg : bool, optional
        Return angle in degrees if True, radians if False (default).

    Returns
    -------
    angle : ndarray or scalar
        The counterclockwise angle from the positive real axis on
        the complex plane, with dtype as numpy.float64.

    See: https://github.com/pydata/xarray/issues/553
         https://github.com/numpy/numpy/blob/v1.13.0/numpy/lib/function_base.py#L2072-L2115
    """
    if deg:
        fact = 180/pi
    else:
        fact = 1.0
    if (issubclass(z.dtype.type, _nx.complexfloating)):
        zimag = z.imag
        zreal = z.real
    else:
        zimag = 0
        zreal = z
    return np.arctan2(zimag, zreal)

[shoyer]

You can also use xarray.ufuncs.angle() for this.

[rth]

@shoyer Aww, great. Thanks for pointing this out.

[NotSqrt]

@shoyer xarray.ufuncs.angle now triggers the DeprecationWarning added in #1962, and numpy/numpy#6266 is still open.

What is the recommended way to keep an DataArray when computing angles ?

Thanks !

[shoyer]

The short answer is that we shouldn't remove xarray.ufuncs.angle until NumPy has a satisfactory overload mechanism for it. This might mean something like http://www.numpy.org/neps/nep-0018-array-function-protocol.html instead.

PendingDeprecationWarning is the weakest form of deprecation warning. Basically it means "We plan to deprecate it in the future, but haven't done so yet."

[NotSqrt]

Thanks !