Closed telegraphic closed 7 years ago
My guess is a precision issue due to pathological input data.
If I replace the input data with this instead:
tmp_arr = np.random.random(size=dx*2).astype(np.float32).view(np.complex64)
then the tests all pass.
Also worth noting that Numpy only supports double-precision FFTs (the output is always complex128
). As an alternative, scipy.fftpack.fft
supports single-precision.
Interesting... OK, closing this for now
This fails:
with this output:
Although TBH it looks like numpy might be wrong?