This is the problem I have. Given a list
xList = [9, 13, 10, 5, 3]
I would like to calculate for sum of each element multiplied by subsequent elements
sum([9*13, 9*10, 9*5 , 9*3]) +
sum([13*10, 13*5, 13*3]) +
sum([10*5, 10*3]) +
sum ([5*3])
in this case the answer is 608 .
Is there a way to do this perhaps with itertools
or natively with numpy
?
Below is a function I came up with. It does the job but it is far from ideal as I would like to add other stuff as well.
def SumProduct(xList):
''' compute the sum of the product
of a list
e.g.
xList = [9, 13, 10, 5, 3]
the result will be
sum([9*13, 9*10, 9*5 , 9*3]) +
sum([13*10, 13*5, 13*3]) +
sum([10*5, 10*3]) +
sum ([5*3])
'''
xSum = 0
for xnr, x in enumerate(xList):
#print xnr, x
xList_1 = np.array(xList[xnr+1:])
#print x * xList_1
xSum = xSum + sum(x * xList_1)
return xSum
Any help appreciated.
NB: In case you wonder, I am trying to implement Krippendorf's alpha with pandas
x = array([9, 13, 10, 5, 3])
result = (x.sum()**2 - x.dot(x)) / 2
This takes advantage of some mathematical simplifications to work in linear time and constant space, compared to other solutions that might have quadratic performance.
Here's a diagram of how this works. Suppose x = array([2, 3, 1])
. Then if you view the products as the areas of rectangles:
x is this stick: -- --- -
x.sum()**2 is this rectangle:
-- --- -
|xx xxx x
|xx xxx x
|xx xxx x
|xx xxx x
|xx xxx x
|xx xxx x
x.dot(x) is this diagonal bit:
-- --- -
|xx
|xx
| xxx
| xxx
| xxx
| x
(x.sum()**2 - x.dot(x)) is the non-diagonal parts:
-- --- -
| xxx x
| xxx x
|xx x
|xx x
|xx x
|xx xxx
and (x.sum()**2 - x.dot(x)) / 2 is the product you want:
-- --- -
| xxx x
| xxx x
| x
| x
| x
|
You actually want combinations not product:
from itertools import combinations
print(sum(a*b for a,b in combinations(xList,2)))
608
Even creating a numpy array from a python list, @user2357112 answer wipes the floor with the rest of us.
In [38]: timeit sum(a*b for a,b in combinations(xlist,2))
10000 loops, best of 3:
89.7 µs per loop
In [40]: timeit sum(mul(*t) for t in itertools.combinations(xlist, 2))
1000 loops, best of 3:
165 µs per loop
In [41]: %%timeit
x = array(arr)
(x.sum()**2 - (x**2).sum()) / 2
....:
100000 loops, best of 3:
10.9 µs per loop
In [42]: timeit np.triu(np.outer(x, x), k=1).sum()
10000 loops, best of 3:
48.1 µs per loop
In [59]: %%timeit
....: xarr = np.array(xList)
....: N = xarr.size
....: range1 = np.arange(N)
....: mask = range1[:,None] < range1
....: out = ((mask*xarr)*xarr[:,None]).sum()
10000 loops, best of 3: 30.4 µs per loop
All the lists/arrays had 50 elements.
Stealing the logic from user2357112 and using it on a normal list with sum python is pretty darn efficient:
In [63]: timeit result = (sum(xList)**2 - sum(x ** 2 for x in xList)) / 2
100000 loops, best of 3:
4.63 µs per loop
But for a large array the numpy solution is still significantly faster.
Here's one way:
In [14]: x = [9, 13, 10, 5, 3]
In [15]: np.triu(np.outer(x, x), k=1).sum()
Out[15]: 608
but I'd go with @user2357112's answer.
It looks like you want to get every combination of two elements (pairs) in that list, compute the product of each pair, and sum over these products:
import itertools
xlist = [9, 13, 10, 5, 3]
pairs = itertools.combinations(xlist, 2)
answer = 0
for pair in pairs:
answer += pair[0] * pair[1]
The one-liner to do this:
import itertools
import operator
sum(operator.mul(*t) for t in itertools.combinations(xlist, 2))
One approach -
xarr = np.array(xList)
N = xarr.size
range1 = np.arange(N)
mask = range1[:,None] < range1
out = ((mask*xarr)*xarr[:,None]).sum()
Another one -
xarr = np.array(xList)
N = xarr.size
range1 = np.arange(N)
R,C = np.where(range1[:,None] < range1)
out = (xarr[R]*xarr[C]).sum()
If you are interested in doing this by hand (without help from the stdlib):
def combinations(L):
for i,elem in enumerate(L):
for e in L[i+1:]:
yield (elem, e)
def main(xlist):
answer = 0
for a,b in combinations(xlist):
answer += a*b
return answer
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.