CODEWITHSUNDEEP
pythonpython-3.xscikit-learn
Laurinda Souza
Laurinda Souza

Reputation: 1337

How to do inverse_transform in OneHotEncoder and LabelEncoder?

I checked that OneHotEncoder does not have an inverse_transform() method. How to get the values ​​back by reversing the transformation?

Code:

from sklearn.preprocessing import LabelEncoder, OneHotEncoder
from sklearn.compose import ColumnTransformer


base = pd.read_csv(caminho + "risco_credito.csv")

previsores = base.loc[:,["historia","divida","garantias","renda"]].values

classe = base.loc[:,"risco"].values


labelencorder_classe = LabelEncoder()
classe_enc = labelencorder_classe.fit_transform(classe)


onehotencorder = ColumnTransformer(transformers=[("OneHot", OneHotEncoder(), [0,1,2,3])],remainder='passthrough')

previsores_enc = onehotencorder.fit_transform(previsores)

For example: Classe_enc and previsores_enc , how to do the inverse transformation, that is, get the values ​​back by reversing the transformation?

Upvotes: 1

Views: 2379

Answers (2)

Venkatachalam
Venkatachalam

Reputation: 16966

May be you are referring to a wrong documentation. The inverse_transform is available for OneHotEncoder and ordinalEncoder.

See here

>>> import numpy as np
>>> from sklearn.preprocessing import OneHotEncoder
>>> from sklearn.compose import ColumnTransformer
>>> x = np.random.choice(['orange','apple', 'mango'],size=(3,1))
>>> ct = ColumnTransformer(transformers=[("OneHot", OneHotEncoder(sparse=False), [0])],
                           remainder='passthrough')

>>> x_trans_ = ct.fit_transform(x)
>>> ct.named_transformers_['OneHot'].inverse_transform(x_trans_)
array([['orange'],
       ['orange'],
       ['apple']], dtype='<U6')

Similarly, you can do it for ordinalEncoder, refer here

Upvotes: 1

ClearStream
ClearStream

Reputation: 31

You could use the argmax function from numpy to get the index of the array element with the maximum value (this will be the index of the element that is 1 since the other values should be 0). Then you could use a dictionary to associate this index with a class label if necessary.

Upvotes: 1

Related Questions