How to use OrdinalEncoder() to set custom order?

Question

I have a column in my Used cars price prediction dataset named "Owner_Type". It has four unique values which are ['First', 'Second', 'Third', 'Fourth']. Now the order that makes the most sense is First > Second > Third > Fourth as the price decreases with respect to this order. How can I give this order to the values with OrdinalEncoder()? Please help me, thank you!

Dubitable

Please provide enough code so others can better understand or reproduce the problem. – Transonic

Answer 1

OrdinalEncoder has a categories parameter which accepts a list of arrays of categories. Here is a code example:

from sklearn.preprocessing import OrdinalEncoder
enc = OrdinalEncoder(categories=[['first','second','third','forth']])
X = [['third'], ['second'], ['first']]
enc.fit(X)
print(enc.transform([['second'], ['first'], ['third'],['forth']]))

Erythritol

Why do you have to fit the encoder to X if you are already supplying the categories? – Verticillate

if you apply transform without applying fit you get an error "AttributeError: 'OrdinalEncoder' object has no attribute 'categories_'" so you must aplly fit_transform or fit the model first and then apply transform. – Erythritol

I see, you're totally right. It's a bit silly, though, since the calling fit when the categories are already supplied does nothing other than create the attribute categories_. It seems that this attribute should be created automatically if categories are supplied. – Verticillate

I'm still struggling with the "correct" order of the categories parameter. In my opinion the order should be reversed, because "First" is more important. What do you think? OrdinalEncoder(categories=[['forth', 'third', 'second', 'first']]) – Marva

Yes you have a point, I think that when we a train a machine learning model or when we use this feature it's better to try both orders and see which one help the model to generalize more. – Erythritol