Thread by @rohanpaul_ai

1/ Why do we need the bias term in ML algorithms such as linear regression and neural networks ? - A thread

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics #programming #ArtificialIntelligence #Data #pythoncode #AI #Stats #DeepLearning #100DaysOfCode

2/ In linear regression, without the bias term your solution has to go through the origin. That is, when all of your features are zero, your predicted value would also have to be zero.

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics #programming

3/ However that may not be the attributes of the training dataset.

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics #programming #ArtificialIntelligence #Data #pythoncode #AI #Stats #DeepLearning #100DaysOfCode #Database #BigData #DataAnalytics

4/ Hence, Adding a bias weight that does not depend on any of the features allows the hyperplane desbribed by your learned weights to more easily fit data that doesn't pass through the origin.

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics

5/ Think of neural network as a good function approximator. In the case of classification, we are interested in approximating a decision boundary in form of a hyperplane.

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics #programming

6/ In case of regression, we want to approximate a function that defines a relationship between X and y.

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics #programming #ArtificialIntelligence #Data #pythoncode #AI #Stats #DeepLearning #100DaysOfCode

7/ If the expected value of your outcome variable is not 0, then we would also need to estimate the bias weight (i.e., the shift from 0) along with the feature weights

#DataScience #MachineLearning #100DaysOfMLCode #DataScientist #Statistics #programming #ArtificialIntelligence