>>> import spellsuggest
>>> c = spellsuggest.Corrector()
>>> c.train('./big.txt') # takes a while; can pass in more than one path.
>>> c.suggest('farke')
[('farce', 1), ('fare', 7)]
>>> c.suggest('ak')
[('ak', 3)]
>>> c.suggest('ake')
[('lake', 27), ('jake', 1), ('rake', 4), ('bake', 2), ('sake', 69), ('cake', 5), ('make', 494), ('take', 590), ('wake', 27)]
>>> c.suggest('lys')
[('lysol', 1)]
>>> c.suggest('lys', 3)
[('lysol', 1), ('ulysses', 3), ('analyse', 1), ('analyst', 1)]
>>> c.correct('sug')
'sugar'
>>> c.correct('sug', 10)
'suggested'
Uses the n-grams partitioning method to index training text. Faaar from anything near perfect. I'm improv[is]ing on it. Obviously, result quality varies directly as corpus text quality.