Check it out on github Last updated: 14/04/2020 03:28:49

Label Encoding .

importance: Machine learning models work on mathematical functions. Mathematical functions don't understand strings. They should be numeric to be added or subtracted. Label Encoding is process of encoding strings or any type to Numbers.

This is alternative to Label encoder in sklearn.preprocessing.LabelEncoder

In [126]:

from sklearn.preprocessing import LabelEncoder

X = {'countries':['india','nepal','china'],'a':[134,4422,333],'b':[22,33,44]}
sdf = pd.DataFrame(X)
sdf.head()

Out[126]:

	countries	a	b
0	india	134	22
1	nepal	4422	33
2	china	333	44

In [127]:

labelencoder_X = LabelEncoder()

Dataset has to be converted into n-dimensional array First.

In [128]:

sdf = sdf.values

In [129]:

sdf

Out[129]:

array([['india', 134, 22],
       ['nepal', 4422, 33],
       ['china', 333, 44]], dtype=object)

In [133]:

sdf[:,0] = labelencoder_X.fit_transform(sdf[:,0])

converting into dataframe again

In [135]:

X = pd.DataFrame(sdf)
X

Out[135]:

	0	1	2
0	1	134	22
1	2	4422	33
2	0	333	44

In [24]:

import pandas as pd
import numpy as np

In [2]:

bridge_type= "Arch cantilever suspended Truss Beam".split()

In [3]:

df = pd.DataFrame(bridge_type,columns=['bridge_type'])

In [4]:

df.head()

Out[4]:

	bridge_type
0	Arch
1	cantilever
2	suspended
3	Truss
4	Beam

converting bridgetype to category and creating new column to store that encoded variables

In [18]:

df[list(df.columns)[0]] = df[list(df.columns)[0]].astype('category')

In [76]:

df['bridge_type_encoded']= df['bridge_type'].cat.codes

Since we don't want our model to understand that Beam is better than arch or suspended is better than cantilever. We might want to dummy them. Dummy in context that : instead of focusing on its precedence(1 , 2 ,3 ...) , convert them in such a way that they all carry same weight for our training model. i.e Arch is as good as Beam

In [28]:

vectored_X = pd.get_dummies(df['bridge_type_encoded'])
vectored_X

Out[28]:

	0	1	2	3	4
0	1	0	0	0	0
1	0	0	0	1	0
2	0	0	0	0	1
3	0	0	1	0	0
4	0	1	0	0	0

We might want to concatenate dummied table and our original Table . lets copy df and drop The strings

In [36]:

df_copy = df.copy()

In [37]:

df_copy = df_copy.drop('bridge_type',axis='columns')

In [38]:

df_copy.head()

Out[38]:

	bridge_type_encoded
0	0
1	3
2	4
3	2
4	1

In [39]:

df_cleaned = pd.concat([vectored_X,df_copy],axis='columns')

In [40]:

df_cleaned.head()

Out[40]:

	0	1	2	3	4	bridge_type_encoded
0	1	0	0	0	0	0
1	0	0	0	1	0	3
2	0	0	0	0	1	4
3	0	0	1	0	0	2
4	0	1	0	0	0	1

Another case. Maybe we don't want our categorical data to be dummied. Let's say they carry some significance. If our data was related to heights [short,medium,tall]. Then they would carry some significance. Short can be thought as 0 and tall can be thought as 2.

In [45]:

heights="Short Medium Tall".split()

In [72]:

data={'Name':['Ashish','Anish','Milan','Gaurav','Nirmal','Avash','Ram'],'height':['short','medium','medium','Girrafe','tall','short','medium']}

In [73]:

df_people = pd.DataFrame(data)

In [74]:

df_height_categorize = pd.Categorical(df_people['height'].values,categories=['short','medium','tall'],ordered=True)

In [75]:

df_height_categorize

Out[75]:

[short, medium, medium, NaN, tall, short, medium]
Categories (3, object): [short < medium < tall]

In [84]:

df_height_categorize = pd.Series(df_height_categorize)

In [87]:

df_people['comprehendable Heights'] = df_height_categorize.cat.codes

In [88]:

df_people

Out[88]:

	Name	height	comprehendable Heights
0	Ashish	short	0
1	Anish	medium	1
2	Milan	medium	1
3	Gaurav	Girrafe	-1
4	Nirmal	tall	2
5	Avash	short	0
6	Ram	medium	1

In [ ]:

Label Encoding in Pandas

Leave a Comment Cancel reply