aggregate - check performance of scipy ndimage

[d1manson]

scipy's ndimage has a few routines that are very similar to specific funcs in aggregate. In particular, extrema may deal with max min in the way that is needed here.

I've not benchmarked it.

[ml31415]

Just had a quick benchmark:

import numpy as np
from scipy.ndimage.measurements import maximum
from numpy_groupies import aggregate

rnd = np.random.RandomState(seed=100)
group_idx = np.repeat(np.arange(1000), 2)
rnd.shuffle(group_idx)
group_idx = np.repeat(group_idx, 10)
a = rnd.randn(group_idx.size)

%timeit maximum(a, labels=group_idx, index=np.unique(group_idx))
# 1000 loops, best of 3: 1.76 ms per loop

%timeit aggregate(group_idx, a, 'max')
# 10000 loops, best of 3: 87.1 µs per loop

# And even without the unique, no big difference
%timeit np.unique(group_idx)
# 1000 loops, best of 3: 407 µs per loop

So, we can still sleep well having the fastest implementation around. :p