Scribd
Upload
37 views

Feature+Encoding

Uploaded by

VAIBHAV PATIL
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
37 views

Feature+Encoding

Uploaded by

VAIBHAV PATIL
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

Feature Encoding

Another very important check before feeding data to Machine Learning Algorithms is to check
whether all the features are of Numerical data types or not. If the dataset has categorical features
with String data types, then the machine cannot interpret it directly, thus it needs to be converted
into numerical fields. This process of converting the String (or object) Data type features to
numerical data type features is termed as encoding.

Categorical features are classified into two categories –


1. Nominal categorical variables
Variables that are just names and are used for labeling/distinction between multiple
classes. In addition, there does not exist any order among them, and none of them have
any numerical significance.
Few examples could be Cities, Gender, Claim (Yes/No) etc.

2. Ordinal categorical variables


Variables in which the value of the data is captured from an ordered set i.e. there exist an
inherent order among the elements.
[email protected]
18XHT46RCY For example: Customer feedback survey on a Likert scale with different levels such as
Poor, Fair, Good, and Very Good.

Once we classify the categorical features present in our dataset, next step is to choose an
appropriate encoding technique.

Three most commonly used Encoding techniques -


1. One-Hot Encoding :
In this technique, for each category of a feature, we create a new column (sometimes
called a dummy variable) with binary encoding (0 or 1) to denote whether a particular
row belongs to this category.
For example, a feature with name of the States would be encoded as below –

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.

This file is meant for personal use by [email protected] only.


Sharing or publishing the contents in part or full is liable for legal action.
A potential drawback of this method is a significant increase in the dimensionality of the
dataset (which is called a Curse of Dimensionality). Thus, one-hot encoding is mostly used
for features with low number of unique values.

In addition, creating dummy variables introduces a form of redundancy to the dataset. If


a feature has six categories like in the example shown above, we only need to have five
dummy variables because, if an observation is neither of the five, it must be the sixth one.
If the redundant dummy variable is not removed, it leads to Multicollinearity issue in the
newly created dummy variables i.e. the independent variables have a relationship among
themselves. Presence of Multicollinearity fails the assumptions of some of the parametric
models such as Linear Regression and Logistic Regression, leading to statistically
insignificant results.

2. Label Encoding:
In label encoding, we replace the categorical value with a numeric value between 0 and
the number of classes minus 1. If the categorical variable value contains 5 distinct classes,
we use (0, 1, 2, 3, and 4) as the labels.
Using the same example as before, below is what the State feature would encoded into
using label encoding –
[email protected]
18XHT46RCY

The labels (0 to 5) are assigned in the Alphabetical order of the unique values i.e. Delhi:0,
Gujrat:1 and so on.

3. Ordinal Encoding
An Ordinal Encoder is used to encode categorical features into an ordinal numerical value
(ordered set). This approach transforms categorical value to numerical value in ordered
sets. This encoding technique appears almost similar to Label Encoding. However, label
encoding would not consider whether a variable is ordinal or not, but in the case of ordinal
encoding, it will assign a sequence of numerical values as per the order of data.

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.

This file is meant for personal use by [email protected] only.


Sharing or publishing the contents in part or full is liable for legal action.
For example, for a typical feedback feature the results of the Ordinal encoding would be
as follows –

Note that if were to use Label Encoding instead, Excellent would assigned 1, Good 2 and
so on, leading to misinformation regarding the order.

Python Codes -

1. One-hot Encoding :

i) Importing the encoders from sklearn

from sklearn.preprocessing import OneHotEncoder


[email protected]
18XHT46RCY
# Initializing the encoder object
# drop = ‘first’ is used to handle the Multicollinearity issue
# sparse = False is used to ensure that the data is not stored in a Sparse matrix format.
one_hot = OneHotEncoder(drop = 'first', sparse=False)
# Fitting the encoder object on the train set
one_hot.fit(data_train[['referral_preferred_city']])

# Transforming the Train and Test sets to be used for modelling


data_train_onehot = one_hot.transform(data_train[['referral_preferred_city']])
data_test_onehot = one_hot.transform(data_test[['referral_preferred_city']])

#Inverse Transforming (Decoding) to get original labels


one_hot.inverse_transform([[1,0,0],[0,1,0],[0,0,1],[0,0,0]])

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.

This file is meant for personal use by [email protected] only.


Sharing or publishing the contents in part or full is liable for legal action.
ii) Using Pandas Get_dummies –
# Argument drop_first = True to handle the issue of Multicollinearity
data_one_hot = pd.get_dummies(data['referral_preferred_city'],drop_first=True)

2. Label Encoding –

# Argument Categories = None to use the unique classes from the feature, and
ordered = None to ensure nominal encoding.
data['ref_city_cat'] = pd.Categorical(values =data['referral_preferred_city'] ,
categories=None, ordered=None)

3. Ordinal Encoding –

[email protected]
i) Importing the encoders from sklearn
18XHT46RCY
from sklearn.preprocessing import OrdinalEncoder

# Initializing the encoder object


# Passing the predefined categories as argument to ensure a required order
ord_encoder = OrdinalEncoder(categories=[['Pune','Delhi','Mumbai','Bangalore']])

# Fitting the encoder object on the train set

ord_encoder.fit(data_train[['referral_preferred_city']])

# Transforming the Train and Test sets to be used for modelling

data_train ['referral_preferred_city_enc'] = ord_encoder.transform(data_train


[['referral_preferred_city']]) .reshape(1,-1)[0]

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.

This file is meant for personal use by [email protected] only.


Sharing or publishing the contents in part or full is liable for legal action.
data_test ['referral_preferred_city_enc'] = ord_encoder.transform(data_test
[['referral_preferred_city']]) .reshape(1,-1)[0]

#Inverse Transforming (Decoding) to get original labels


ord_encoder.inverse_transform([[0],[1],[2],[3]])

ii) Using Pandas Categorical function –

#Argument categories to define the custom classes, and ordered = True to


achieve an ordered set (Ordinal Encoding)
data['ref_city_cat']= pd.Categorical(values =data['referral_preferred_city'] ,
categories=['Mumbai','Delhi','Bangalore','Pune'],
ordered=True)

[email protected]
18XHT46RCY

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited.

This file is meant for personal use by [email protected] only.


Sharing or publishing the contents in part or full is liable for legal action.

You might also like