Question 1. [70 marks] This question is designed to help you understand the working of a decision tree (DT). You will find a dataset,here in this link (https://drive.google.com/file/d/12dFVkfPbQXQkVN8ESa7dhZoO3Mnsgwbe/view?usp=sharing ), containing information used to classify penguins into 3 species. You need to implement a classification decision tree (DT) from scratch [you are not permitted to use any 3rd-party librarys function for the classifier e.g. scikit. You may, however, use built-in functions for auxiliary tasks like train/test split, etc.]. With respect to the cost function to be used to find splits, everywhere in this question, you have to implement one of the following based on your roll number: Odd roll numbers - Gini index Even roll numbers - entropy The implementation includes the following tasks. - 1. Perform pre-processing and visualization of the dataset. Perform categorical encoding wherever applicable and split the data into train and test sets - [7 marks]. 2. Implement the cost function as per your roll num. (details above) - [5 marks]. 3. In order for the decision tree to work successfully, continuous variables need to be converted to categorical variables first. To do this, you need to implement a decision function that makes this split. Let us call that cont_to_cat(). The details of the function are the following. [10 marks]:- a. Assume that the continuous variables are independent of each other i.e. assuming 2 continuous variables A and B, the split of A does not in any way affect the split you will perform in B.

b. The continuous variables should only be split into 2 categories, and the optimal split is one that divides the samples the best, based on the value of the function you have been allotted (as per your roll number).

4. After step 2, all the attributes would have categorical values, so now you can go ahead and implement the training function. This would include implementing the following helper functions: [25 marks] a. Get the attribute that leads to the best split b. Make that split c. Repeat these steps for the newly-created split 5. The DT should also include the following properties in the train function a. There should be a max depth that should be defined i.e. a depth after which the tree shouldnt be allowed to grow [5 marks] b. The algorithm should self-identify when there is no information gain being done, i.e. the model has plateaued in its training and shouldnt grow further. [5 marks] 6. Write a function which is responsible for classification (i.e. at test time). [10 marks] 7. Find out the accuracy you get on the test data (overall and class-wise). [3 marks] Note: Implementing the wrong cost function will attract a loss of credit in the above question.