The problem has 3 classes and 4 input features. The possible classes are (B: balanced, L: left, R: right), and the 4 input features are: (LW: left weight, RW: right weight, LD: left distance, RD: Right distance). The balance shows up the value B when LW + LD = RD + RW. Each raw of the file data is constructed as the following: Class, LW, LD, RW, RD. Example: B, 3, 2, 4, 1. In addition, all the input features might have values 1, ..., 5. For more information refer to website given before. Remark: you are free to propose a preprocessing of the input features, also youre advised to try different values of the learning rate and observe the training curve. Remark: in your code, you are asked to use python, numpy and matplotlib. You shouldnt use libraries like scikit, tensorflow, pytorch, or any other library dedicated to AI. Remark: Provide an intelligible and commented algorithm that shows your deep understanding of your code. Remark: as far as the normalized algorithm is concerned, it is usual to prevent a possible division by zero, so a small real positive number is added to the divisor.

Exercise 3 (BONUS): In this question you are free to propose the dataset of your choice and apply the Logistic classification algorithm to train either a full single layer or a single perceptron depending on the data set. Provide the figures you obtain from the code