Randomly select list from list of lists in python depending on weights

Question

I have a list of lists, where each list is associated with a score/weight. I want to produce a new list of lists by randomly selecting from the first one so that those with higher scores will appear more often. The line below works fine when population is just a normal list. But I want to have it for a list of lists.

population = [['a','b'],['b','a'],['c','b']]
list_of_prob = [0.2, 0.2, 0.6]

population = np.random.choice(population, 10, replace=True, p=list_of_prob)

This will give the output ValueError: a must be 1-dimensional

@Chicony you should mention in the question that it's not necessary should be `numpy` — vishes_shell, Oct 11 '16 at 09:06

RemcoGerlich · Accepted Answer · 2016-10-11T08:41:51.887

9

Instead of passing the actual list, pass a list with indexes into the list.

np.random.choice already allows this, if you pass an int n then it works as if you passed np.arange(n).

So

choice_indices = np.random.choice(len(population), 10, replace=True, p=list_of_prob)
choices = [population[i] for i in choice_indices]

edited Oct 11 '16 at 08:41

answered Oct 11 '16 at 08:34

RemcoGerlich

30,470
6
61
79

`TypeError: only integer arrays with one element can be converted to an index` – roganjosh Oct 11 '16 at 08:37
@roganjosh use `range(len(population))` – Sardorbek Imomaliev Oct 11 '16 at 08:38
2

This will actually work like a charm, just if you do `choice = [population[i] for i in choice_index]`, instead of `choice = population[choice_index]` as `choice_index` is a `list` – lycuid Oct 11 '16 at 08:39
@SardorbekImomaliev still the same error. It comes from line `choice = population[choice_index]` – roganjosh Oct 11 '16 at 08:39
@roganjosh: sorry, I didn't notice you were getting 10 values, assumed 1. Will edit. – RemcoGerlich Oct 11 '16 at 08:40
Now it's fixed :) – roganjosh Oct 11 '16 at 08:43

Randomly select list from list of lists in python depending on weights

1 Answers1