How To Implement Weighted Mean Square Error in Python?

In this article, we discussed the implementation of weighted mean square error using python.

Mean squared error is a vital statistical concept, that is nowadays widely used in Machine learning and Deep learning algorithm. Mean squared error is basically a measure of the average squared difference between the estimated values and the actual value. It is also called a mean squared deviation and is most of the time used to calibrate the accuracy of the predicted output. In this article, let us discuss a variety of mean squared errors called weighted mean square errors. 

Weighted mean square error enables to provide more importance or additional weightage for a particular set of points (points of interest) when compared to others. When handling imbalanced data, a weighted mean square error can be a vital performance metric. Python provides a wide variety of packages to implement mean squared and weighted mean square at one go, here we can make use of simple functions to implement weighted mean squared error.

Formula to calculate the weighted mean square error:

Implementation of Weighted Mean Square Error

• For demonstration purposes let us create a sample data frame, with augmented actual and predicted values, as shown.
• Calculate the squared difference between actual and predicted values.
• Define the weights for each data point based on the importance
• Now, use the weights  to calculate the weighted mean square error as shown

Code Implementation:

import pandas as pd
import numpy as np
import random

# create a dataset with actual and
# predicted values
d = {'Actual': np.arange(0, 20, 2)*np.sin(2),
     'Predicted': np.arange(0, 20, 2)*np.cos(2)}

# convert the data to pandas dataframe
data = pd.DataFrame(data=d)

# create a weights array based on 
# the importance
y_weights = np.arange(2, 4, 0.2)

# calculate the squared difference
diff = (data['Actual']-data['Predicted'])**2

# compute the weighted mean square error
weighted_mean_sq_error = np.sum(diff * y_weights) / np.sum(y_weights)

Output:

Let us cross verify the result with the result of the scikit-learn package. to verify the correctness,

Code:

# compare the results with sklearn package
weighted_mean_sq_error_sklearn = np.average(
    (data['Actual']-data['Predicted'])**2, axis=0, weights=y_weights)

weighted_mean_sq_error_sklearn

Output: