Raveling coefficients example doesn't work in 2D

[zaccharieramzi]

Hi,

I tried the example given in the docs for raveling coefficients in n-D: https://pywavelets.readthedocs.io/en/latest/ref/dwt-coefficient-handling.html#raveling-and-unraveling-coefficients-to-from-a-1d-array .

It works perfectly. However, if I chose to use wavedec2 instead of wavedecn, it fails:

import pywt
from numpy.testing import assert_array_almost_equal
cam = pywt.data.camera()
coeffs = pywt.wavedec2(cam, wavelet='db2', level=3)
arr, coeff_slices, coeff_shapes = pywt.ravel_coeffs(coeffs)
coeffs_from_arr = pywt.unravel_coeffs(arr, coeff_slices, coeff_shapes)
cam_recon = pywt.waverec2(coeffs_from_arr, wavelet='db2')
assert_array_almost_equal(cam, cam_recon)

gives:

ValueError                                Traceback (most recent call last)
<ipython-input-187-3b8a21e194b7> in <module>
      4 arr, coeff_slices, coeff_shapes = pywt.ravel_coeffs(coeffs)
      5 coeffs_from_arr = pywt.unravel_coeffs(arr, coeff_slices, coeff_shapes)
----> 6 cam_recon = pywt.waverec2(coeffs_from_arr, wavelet='db2')
      7 assert_array_almost_equal(cam, cam_recon)

~/workspace/pysap/venv/lib/python3.5/site-packages/PyWavelets-1.0.1-py3.5-linux-x86_64.egg/pywt/_multilevel.py in waverec2(coeffs, wavelet, mode, axes)
    323             idxs = tuple(slice(None, -1 if a_len == d_len + 1 else None)
    324                          for a_len, d_len in zip(a.shape, d_shape))
--> 325         a = idwt2((a[idxs], d), wavelet, mode, axes)
    326 
    327     return a

~/workspace/pysap/venv/lib/python3.5/site-packages/PyWavelets-1.0.1-py3.5-linux-x86_64.egg/pywt/_multidim.py in idwt2(coeffs, wavelet, mode, axes)
    116 
    117     coeffs = {'aa': LL, 'da': HL, 'ad': LH, 'dd': HH}
--> 118     return idwtn(coeffs, wavelet, mode, axes)
    119 
    120 

~/workspace/pysap/venv/lib/python3.5/site-packages/PyWavelets-1.0.1-py3.5-linux-x86_64.egg/pywt/_multidim.py in idwtn(coeffs, wavelet, mode, axes)
    275                          "band")
    276     if any(s != coeff_shape for s in coeff_shapes):
--> 277         raise ValueError("`coeffs` must all be of equal size (or None)")
    278 
    279     if axes is None:

ValueError: `coeffs` must all be of equal size (or None)

After looking into the unraveled coeffs, it appears that some of them are actually dictionary rather than tuple. Example: type(coeffs[1]) == tuple and type(coeffs_from_arr[1]) == dict. I haven't looked much more into it, so it might not be the cause of this, but I guess for the sake of idempotence, this shouldn't be the case. Also it's not the case in n-D.

[grlee77]

In this case, you need to specify the output_format argument to pywt.unravel_coeffs as follows:

coeffs_from_arr = pywt.unravel_coeffs(arr, coeff_slices, coeff_shapes, output_format='wavedec2')

wavedecn is the default. Perhaps for the example in the documentation we should explicitly specify output_format='wavedecn' to make it less likely for users to overlook the option?

[grlee77]

We could also potentially add some sort of check in waverec2 to raise a more informative error if a dictionary is encountered unexpectedly.

[zaccharieramzi]

It works indeed.
I can try and add that in the docs, as well as the check in waverec2 if you want me to.

[grlee77]

I can try and add that in the docs, as well as the check in waverec2 if you want me to.

I just submitted #471 to do this. My apologies if you had already started working on this (I just now saw your reply).