bla.py

from sklearn.preprocessing import LabelEncoder
import pandas as pd
import bisect

le = LabelEncoder()
train_c = le.fit_transform(['a', 'b', 'c', 'a'])
test_c = pd.Series(['a', 'b', 'c', 'd']).map(
    lambda s: '<unk>' if s not in le.classes_ else s
)

print(test_c.tolist()) # ['a', 'b', 'c', '<unk>']

# We need to insert <unk> in the sorted classes order, because
# le.transform uses np.searchsorted:
# https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/preprocessing/label.py#L149
# However, this feels really clunky!

le_classes = le.classes_.tolist()
bisect.insort_left(le_classes, '<unk>')
le.classes_ = le_classes

print(le.transform(test_c)) # [1, 2, 3, 0]