reconstruction from array_to_coeff and waverec

[SylvainLan]

Hello,

I am getting an error while running the following code

x = np.random.randn(256)
w = pywt.wavedec(x, wavelet='haar', level=3)
w_arr, w_cs = pywt.coeffs_to_array(w)
w2 = pywt.array_to_coeffs(w_arr, w_cs)
x_recon = pywt.waverec(w2, wavelet='haar')

The error message is

~/anaconda3/lib/python3.6/site-packages/pywt/_multilevel.py in waverec(coeffs, wavelet, mode, axis)
    145         if (a is not None) and (d is not None):
    146             try:
--> 147                 if a.shape[axis] == d.shape[axis] + 1:
    148                     a = a[[slice(s) for s in d.shape]]
    149                 elif a.shape[axis] != d.shape[axis]:

AttributeError: 'dict' object has no attribute 'shape'

The code runs if I change waverec to waverecn however.

I can try to look at this soon but I'm new to pywavelet so I don't know how long it will take me !

Syl

[grlee77]

array_to_coeffs defaults to returning coeficients compatibile with wavedecn. For use with waverec you have to provide an additional argument:

w2 = pywt.array_to_coeffs(w_arr, w_cs, output_format='wavedec')

The reason for this is that it is not possible to tell from the concatenated array which transform generated the data. Perhaps it would have been better to make output_format a required argument (although at this point I would vote to leave it as is for backward compatibility)?

[grlee77]

This is not a bug, but it is perhaps worth adding a check in wavedec itself to return a more informative error message if the coefficients received are not of the proper format. As is, the attribute error that occurred in your case is not helpful in diagnosing the problem.

[SylvainLan]

I changed the error message in waverec, is that what you meant ?