Column Transformer in Machine Learning

Column Tranfsformer is used in data preprocessing. It creates pipelines and transforms row in one go. Without it we need to extract the columns from dataframe , then transform it , and ultimately we need to further concatenate all such transformed columns ( which are now in form of numpy arrays) . Extracting , transforming , concatenating and converting into dataframe is a laborious task.

Get Animesh Tripathi’s stories in your inbox
Join Medium for free to get updates from this writer.
Remember me for faster sign in
#import col transf class
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OrdinalEncoder,SimpleImputer
#create its instance
transformer = ColumnTransformer( transformers=[
            ('trnf1',OrdinalEncoder(categories([['bad','average','poor'],['school','UG','PG']])),['grade','education'])
            ('trnf2',SimpleImputer(),['Age'])
],remainders='passthrough')

transformer.fit_transform(X_train)
transformer.transform(X_test) 

Step By Step Explaination

First we need to import classes that are to be used like column transformer , Ordinal encoder and simple imputer .
Then we create instance , there are two parameters used i.e. transformer and remainder , former tells which columns are to be tranformer and later shows what to do with leftover columns . Default value of remainder is ‘drop’ which drops the remaining columns . If we don’t want to drop and keep them as they are we give value ‘passthrough’ to the remainder.
Each value is a tuple in transformer , ‘tnfr1’ is just a label to that particular transformation , ‘OrdinalEncoding’ it tells which tranformation to apply , here it means categorical values are given numerical value based on order which they are written i.e. ‘bad’->0,’average’->1,’poor’->2 , and the last is the column name, for which whole story of tuple is written .