[FIXED] Scikit learn preprocessing cannot understand the output using min_frequency argument in OneHotencoder class

Issue

Consider the below array t. When using min_frequency kwarg in the OneHotEncoder class, I cannot understand why the category snake is still present when transforming a new array. There are 2/40 events of this label. Should the shape of e be (4,3) instead?

sklearn.__version__ == '1.1.1'

t = np.array([['dog'] * 8 + ['cat'] * 20 + ['rabbit'] * 10 +
               ['snake'] * 2], dtype=object).T
enc = OneHotEncoder(min_frequency= 4/40, 
                    sparse=False).fit(t)

print(enc.infrequent_categories_)
# [array(['snake'], dtype=object)]

e = enc.transform(np.array([['dog'], ['cat'], ['dog'], ['snake']]))
array([[0., 1., 0., 0.],
       [1., 0., 0., 0.],
       [0., 1., 0., 0.],
       [0., 0., 0., 1.]]) # snake is present?

Solution

Check out enc.get_feature_names_out():

array(['x0_cat', 'x0_dog', 'x0_rabbit', 'x0_infrequent_sklearn'],
      dtype=object)

"snake" isn't considered its own category anymore, but lumped into the infrequent category. If you added some other rare categories, they'd be assigned to the same, and if you additionally set handle_unknown="infrequent_if_exist", you would also encode unseen categories to the same.

Answered By - Ben Reiniger