Brute force feature selection - sum items across list combinations-CodePudding

I have x lists of parameters & a list containing a list of the x lists. A total score is calculated by summing each corresponding item in each list.

For example, I have 2 lists & a main list like so:

a = [1,2]
b = [2,1]
mainList = ['a','b']

I want all the unique combinations - this is given by:

for L in range(0, len(mainList) 1):
for subset in itertools.combinations(mainList, L):
    print(subset)

Output:

()
('a',)
('b',)
('a', 'b')

For each permutation, I wish to create a new list containing the overall score by summing the items of each included list.

So for example, in this case the desired outputs would be:

[0,0] 
[1,2] #i.e. just a
[2,1] #i.e. just b
[3,3] #i.e. a b

This is where I'm stuck - I'm not sure how to get the desired output above.

I suspect I'm overcomplicating it & there's just a quick way to do it.

CodePudding user response：

What you need is to somehow connect a & b strings with corresponding lists. I think the cleanest way is to have a dictionary containing lists' names and lists with their values.

Here is an example:

list_dict = {'a': a, 'b': b}

for L in range(0, len(mainList) 1):
    for subset in itertools.combinations(mainList, L):
        temp = [0, 0]
        for sub in subset:
            temp[0]  = list_dict[sub][0]
            temp[1]  = list_dict[sub][1]
        print(temp)

CodePudding user response：

Use zip(*...) to 'transpose' the lists (e.g. convert [[1, 2], [3, 4]] to [(1, 3), (2, 4)]) and then sum the elements:

a = [1, 2]
b = [3, 4]
main = [a, b]  # make main a list of the lists rather than text labels

for length in range(len(main) 1):
    for subset in itertools.combinations(main, length):
        result = [sum(t) for t in zip(*subset)]
        if result:
            print(result)
        else:
            print([0] * len(main[0])

Note however that the first sum (sum of no lists) is [] so we need a specific line to replace that.