I have a numpy array of a certain length, and want to produce a much shorter array, where each value in the shorter array is aggregated based on a bin of values in the original array. In my specific case, this happens to be an audio signal and the function I'm applying is a max over each bin.
It feels like there must be a function like numpy.interp or scipy.interpolate.interp1d which could make this easier. I swear there must be a simpler/cleaner way to do this with a numpy or scipy built-in.
This is the code I write now:
# assume a is an np.array with len(a) == 510
shorten_length = 50
a_shortened = np.zeros(len(a) // shorten_length)
for i in range(len(a_shortened)):
slice_end = min(len(a), (i+1) * shorten_length) - 1
_bin_max = np.max(a[i * shorten_length:slice_end+1])
a_shortened[i] = _bin_max
As long as you are dropping the last slice (when it's smaller than shorten_length),
you can use this approach:
valid_length = len(a) - len(a) % shorten_length
a_shortened = a[:valid_length].reshape(-1, shorten_length).max(axis=1)
No array copy is involved during slicing or reshaping.