CODEWITHSUNDEEP
pythonpandasnumpyrandompython-3.8
Daniel Klöck
Daniel Klöck

Reputation: 21137

select a row index weighted by value

weight dictionary: {1:0.1, 2:0.9} (there is a 10% probability of an item with the value 1 of being selected, 90% with value 2)

exmple value row: [0, 0, 1, 0, 2, 1] (the will only be 0s and values that are contained in the dictionary)

The output should be a randomly chosen index

For the example row, the probabilities of the index of being selected for each item in the row should be [0, 0, 0.05, 0, 0.9, 0.05] (note that since the row contains two different 1 elements, each of them should have a prob of 0.05 of being selected, since the weight counts towards an item with that value being selected)

Upvotes: 2

Views: 85

Answers (1)

Ch3steR
Ch3steR

Reputation: 20669

You can use np.select here.

wt = {1:0.1, 2:0.9}
a = np.array([0, 0, 1, 0, 2, 1])
choicelist = [a==i for i in wt.keys()]
condlist = [v/np.count_nonzero(a==k) for k,v in wt.items()]
np.select(choicelist, condlist)
# array([0.  , 0.  , 0.05, 0.  , 0.9 , 0.05])

Upvotes: 3

Related Questions