[FIXED] Is there a python function similar to sklearn.PolynomialFeatures but for strings?

Issue

The sklearn.PolynomialFeatures function generates the polynomial and interaction features of a vector. For example :

>>> X = [[1,2,3]]
>>> G = sklearn.PolynomialFeatures(degree = 3, interaction_only = True, bias = False)
>>> G.fit_transform(X)
>>> print(G)
>>>
array([[1., 2., 3., 2., 3., 6., 6.]])

Is there an equivalent function that could work for strings so that if the input array is X = [['a','b','c']] the function would output array([['a','b','c','ab','ac','bc','abc']]) and that the function could take any input vector ? If no such function exist, do you have an idea on how to create it ?

Solution

It looks like you're looking for the superset of the input list of strings. This is fairly easy to implement using itertools, though if you want to have the fit/transform structure (allowing you to include the transformer in a pipeline), you can define your own transformer inheriting from TransformerMixin. Otherwise just use the code contained in the transform method:

from sklearn.base import TransformerMixin
from itertools import combinations, chain

class NSuperset(TransformerMixin):
    def __init__(self, n):
        self.n = n

    def fit(self, X):
        return self

    def transform(self, X):
        superset = [[''.join(c) for x in X for c in combinations(x, r=i)] 
                    for i in range(1,self.n+1)]
        return list(chain.from_iterable(superset))

---

ss = NSuperset(n=3)

X = [['a','b','c']]
ss.fit_transform(X)
# ['a', 'b', 'c', 'ab', 'ac', 'bc', 'abc']

Answered By - yatu