Feature scaling

1. Normalization - Changing the values of features in dataset to a range of between 0 and 1.

2. Standardization - Changing mean to 0 and standard deviation to 1 for all features of a given dataset. Range of values can be any lower and upper limit (not necessary to be between 0 and 1).

Scaling is required in case of algorithms which requires gradient descent calculations.

Normalization code:

import pandas as pd
import numpy as np
from sklearn import datasets

# getting the dataset
wine = datasets.load_wine()

# converting data to dataframe
wine_df = pd.DataFrame(wine.data)

# setting the column names
wine_df.columns = wine.feature_names

# By running this code, we can the see the min and max is not ranging between 0 and 1.
wine_df.describe()

# getting only the input columns in numpy array format
X = wine.data

# Applying normalization
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()

scaler.fit(X)
wine_scaled = scaler.transform(X)
wine_scaled # final scaled data

# Converting it to dataframe
wine_scaled_df = pd.DataFrame(wine_scaled)

# setting the column names
wine_scaled_df.columns = wine.feature_names

# After the normalization, after running this code we can that the values are ranging between 0 and 1
wine_scaled_df.describe()

Before normalization -

After normalization -

Standardization code:

import pandas as pd
import numpy as np
from sklearn import datasets

# getting the dataset
wine = datasets.load_wine()

# converting data into dataframe
wine_df = pd.DataFrame(wine.data)

# setting up the column names
wine_df.columns = wine.feature_names

# After running this code we will see that the mean and standard deviation (std) values are not 0 and 1 respectively.
wine_df.describe()

# getting only the input columns in numpy array format
X = wine.data

# Applying standardization
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()

wine_scaled = scaler.fit_transform(X)

# Converting it to dataframe
wine_scaled_df = pd.DataFrame(wine_scaled)

# Setting up the column names
wine_scaled_df.columns = wine.feature_names

# After running this code, we can see that the mean is 0 and std is 0 for all features.
wine_scaled_df.describe().round(2)

Before standardization -

After standardization -