trouble with real=True in dft

[navidcy]

I am having some trouble using the real fft. I have a time series of temperature data:

da = surf_temp_1selection.copy()
months = np.arange(da.time.size)+1
da.time.values = months
da = da.chunk({'time':None})
da = da.mean(dim={'xt_ocean', 'yt_ocean'})
da

<xarray.DataArray 'surface_temp' (time: 2520)>
dask.array<mean_agg-aggregate, shape=(2520,), dtype=float32, chunksize=(2520,), chunktype=numpy.ndarray>
Coordinates:
  * time     (time) int64 1 2 3 4 5 6 7 8 ... 2514 2515 2516 2517 2518 2519 2520

When I use dft without the real=True flag all seems good:

dahat = xrft.dft(da, dim=['time'], window=True)
dahat

<xarray.DataArray 'rechunk-merge-43b3f58ebe71665aef5d349f10b36e8c' (freq_time: 2520)>
dask.array<rechunk-merge, shape=(2520,), dtype=complex128, chunksize=(2520,), chunktype=numpy.ndarray>
Coordinates:
  * freq_time          (freq_time) float64 -0.5 -0.4996 ... 0.4992 0.4996
    freq_time_spacing  float64 0.0003968

But when I give real=True I get:

dahat = xrft.dft(da, dim=['time'], window=True, real=True)
dahat


---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-89-2e819cfb9956> in <module>
      6 
      7 print(da)
----> 8 dahat = xrft.dft(da, dim=['time'], window=True, real=True)
      9 print(" ");print(" ");print(" ");
     10 print(dahat)

/g/data3/hh5/public/apps/miniconda3/envs/analysis3-19.10/lib/python3.6/site-packages/xrft/xrft.py in dft(da, spacing_tol, dim, real, shift, detrend, window, chunks_to_segments)
    250 
    251     rawdims = da.dims
--> 252     da, trans = _transpose(da, real)
    253     if dim is None:
    254         dim = list(da.dims)

/g/data3/hh5/public/apps/miniconda3/envs/analysis3-19.10/lib/python3.6/site-packages/xrft/xrft.py in _transpose(da, real, trans)
    194         transdim = list(da.dims)
    195         if real not in transdim:
--> 196             raise ValueError("The dimension along real FT is taken must "
    197                             "be one of the existing dimensions.")
    198         elif real != transdim[-1]:

ValueError: The dimension along real FT is taken must be one of the existing dimensions.

Am I doing something wrong?

[roxyboy]

real takes the dimension to take the real FFT over so in your case, dahat = xrft.dft(da, window=True, real='time') should do the trick.

[roxyboy]

Did it work @navidcy ? If so, I'm gonna close this :)

[navidcy]

@roxyboy, I got a different error now

xrft.dft(da, dim=['time'], real='time', window=True)

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-17-3ee90d6aade1> in <module>
----> 1 xrft.dft(da, dim=['time'], real='time', window=True)

/g/data3/hh5/public/apps/miniconda3/envs/analysis3-19.07/lib/python3.6/site-packages/xrft/xrft.py in dft(da, spacing_tol, dim, real, shift, detrend, window, chunks_to_segments)
    350     if trans:
    351         enddims = [prefix + d for d in rawdims if d in dim]
--> 352         return daft.transpose(*enddims)
    353     else:
    354         return daft

/g/data3/hh5/public/apps/miniconda3/envs/analysis3-19.07/lib/python3.6/site-packages/xarray/core/dataarray.py in transpose(self, transpose_coords, *dims)
   1850                 raise ValueError(
   1851                     "arguments to transpose (%s) must be "
-> 1852                     "permuted array dimensions (%s)" % (dims, tuple(self.dims))
   1853                 )
   1854 

ValueError: arguments to transpose (('freq_time',)) must be permuted array dimensions (('yt_ocean', 'xt_ocean', 'freq_time'))

But you are right. This doesn't seem to be an issue of the package so I'm closing it and we can continue discussion over email.

[roxyboy]

This is probably the same issue as #78 . It seems that xarray did some updates which we may have to update xrft...

[navidcy]

OK, I'm reopening this then.
I don't really understand the solution @TomNicholas suggests in #78.

How come the tests didn't break when xarray updated? Perhaps you need to make stricter tests? (I'd happily do it but I'm not very good with python...)

[roxyboy]

That's probably because our test environment is Python 3.6.7 while as the new xarray features are in Python 3.7

[roxyboy]

@navidcy Do you wanna give PR #88 a try to see if it works for your needs? I think it should.

[rabernat]

Xarray is the same in Python 3.6 and 3.7

[roxyboy]

Hmm, then I'm not sure why real DFT generated errors in Python 3.7 and not in 3.6. In any case, the tests for real DFT in Python 3.7 all pass in PR #88.