Sum array by number in numpy

This is a vectorized method of doing this sum based on the implementation of numpy.unique. According to my timings it is up to 500 times faster than the loop method and up to 100 times faster than the histogram method.

def sum_by_group(values, groups):
    order = np.argsort(groups)
    groups = groups[order]
    values = values[order]
    values.cumsum(out=values)
    index = np.ones(len(groups), 'bool')
    index[:-1] = groups[1:] != groups[:-1]
    values = values[index]
    groups = groups[index]
    values[1:] = values[1:] - values[:-1]
    return values, groups

The numpy function bincount was made exactly for this purpose and I'm sure it will be much faster than the other methods for all sizes of inputs:

data = [1,2,3,4,5,6]
ids  = [0,0,1,2,2,1]

np.bincount(ids, weights=data) #returns [3,9,9] as a float64 array

The i-th element of the output is the sum of all the data elements corresponding to "id" i.

Hope that helps.

There's more than one way to do this, but here's one way:

import numpy as np
data = np.arange(1, 7)
groups = np.array([0,0,1,2,2,1])

unique_groups = np.unique(groups)
sums = []
for group in unique_groups:
    sums.append(data[groups == group].sum())

You can vectorize things so that there's no for loop at all, but I'd recommend against it. It becomes unreadable, and will require a couple of 2D temporary arrays, which could require large amounts of memory if you have a lot of data.

Edit: Here's one way you could entirely vectorize. Keep in mind that this may (and likely will) be slower than the version above. (And there may be a better way to vectorize this, but it's late and I'm tired, so this is just the first thing to pop into my head...)

However, keep in mind that this is a bad example... You're really better off (both in terms of speed and readability) with the loop above...

import numpy as np
data = np.arange(1, 7)
groups = np.array([0,0,1,2,2,1])

unique_groups = np.unique(groups)

# Forgive the bad naming here...
# I can't think of more descriptive variable names at the moment...
x, y = np.meshgrid(groups, unique_groups)
data_stack = np.tile(data, (unique_groups.size, 1))

data_in_group = np.zeros_like(data_stack)
data_in_group[x==y] = data_stack[x==y]

sums = data_in_group.sum(axis=1)

If the groups are indexed by consecutive integers, you can abuse the numpy.histogram() function to get the result:

data = numpy.arange(1, 7)
groups = numpy.array([0,0,1,2,2,1])
sums = numpy.histogram(groups, 
                       bins=numpy.arange(groups.min(), groups.max()+2), 
                       weights=data)[0]
# array([3, 9, 9])

This will avoid any Python loops.