Scaling numbers in machine learning is a common pre-processing technique to standardize the independent features present in the data in a fixed range. When applied to a Python sequence, such as a Pandas Series, scaling results in a new sequence such that your entire values in a column comes under a range. For example if the range is ( 0 ,1 ) your entire data within that column will be in the range 0,1 only.
Example:
if the sequence is [1, 2, 3] then the scaled sequence is [0, 0.5, 1]
Application:
- In machine learning, scaling can improve the convergence speed of various algorithms.
- Often in machine learning, you will come across data sets with a huge variation, and it will be difficult for many machine learning models well on that data so in that case scaling helps to keep the data within a range.
Note: We will be using Scikit-learn in this article to scale the pandas dataframe.
Steps:
- Import pandas and sklearn library in python.
- Call the DataFrame constructor to return a new DataFrame.
- Create an instance of sklearn.preprocessing.MinMaxScaler.
- Call sklearn.preprocessing.MinMaxScaler.fit_transform(df[[column_name]]) to return the Pandas DataFrame df from the first step with the specified column min-max scaled.
Example 1 :
A very basic example of how MinMax
Python3
# importing the required libraries import pandas as pd from sklearn.preprocessing import MinMaxScaler # creating a dataframe for example pd_data = pd.DataFrame({ "Item": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], "Price": [100, 300, 250, 120, 910, 345, 124, 1000, 289, 500] }) # Creating an instance of the sklearn.preprocessing.MinMaxScaler() scaler = MinMaxScaler() # Scaling the Price column of the created dataFrame and storing # the result in ScaledPrice Column pd_data[["ScaledPrice"]] = scaler.fit_transform(pd_data[["Price"]]) print(pd_data) |
Output :
Example 2 : You can also scale more than one pandas, DataFrame’s column at a time, you just have to pass the column names in the MinMaxScaler.fit_transform() function.
Python3
# importing the required libraries import pandas as pd from sklearn.preprocessing import MinMaxScaler # creating a dataframe for example pd_data = pd.DataFrame({ "Item": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], "Price": [100, 300, 250, 120, 910, 345, 124, 1000, 289, 500], "Weight": [200, 203, 350, 100, 560, 456, 700, 250, 800, 389] }) # Creating an instance of the sklearn.preprocessing.MinMaxScaler() scaler = MinMaxScaler() # Scaling the Price column of the created dataFrame and storing # the result in ScaledPrice Column pd_data[["ScaledPrice", "ScaledWeight"]] = scaler.fit_transform( pd_data[["Price", "Weight"]]) print(pd_data) |
Output :
Example 3: By default, the scale value used the class MinMaxScaler() is (0,1) but you can change it to any value you want as per your need.
Python3
# importing the required libraries import pandas as pd from sklearn.preprocessing import MinMaxScaler # creating a dataframe for example pd_data = pd.DataFrame({ "Item": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], "Price": [100, 300, 250, 120, 910, 345, 124, 1000, 289, 500] }) # Creating an instance of the sklearn.preprocessing.MinMaxScaler() # specifying the min and max value of the scale scaler = MinMaxScaler(feature_range=(20, 500)) # Scaling the Price column of the created dataFrame # and storing the result in ScaledPrice Column pd_data[["ScaledPrice"]] = scaler.fit_transform(pd_data[["Price"]]) print(pd_data) |
Output :
