IITM BS Machine Learning Courses: MLF, MLT and MLP Explained

The IITM BS Diploma in Data Science has three machine learning courses, and all three are mandatory. Machine Learning Foundations (MLF) teaches the maths behind ML. Machine Learning Techniques (MLT) teaches the main methods. Machine Learning Practice (MLP) puts those methods to work in Python with scikit-learn, and it comes with a project run on Kaggle. The best preparation is a solid grip on Maths 2 and Stats 2 before you start.

The three courses at a glance

So MLF and MLT can run in the same term, or MLF can come first. MLP needs both done. The MLP project runs alongside MLP or after it. Like every diploma course, they open only after you pass all 8 foundation courses. The full diploma structure is in the Diploma in Data Science post.

What MLF covers

MLF says it lays the groundwork for the later ML courses. Its weeks, from the course page:

• Week 1: what makes a problem an ML problem.
• Week 2: calculus.
• Weeks 3 to 6: least squares regression, eigenvalues and eigenvectors, symmetric matrices, singular value decomposition and PCA.
• Weeks 7 to 9: unconstrained and constrained optimisation, convex functions, Lagrange multipliers, logistic regression as optimisation.
• Weeks 10 to 12: probabilistic models, the exponential family, parameter estimation and expectation maximisation.

This is Maths 2 and Stats 2, pointed at ML. If rank, basis, inner products or partial derivatives feel shaky, revise them before week 3.

What MLT covers

MLT goes through the main methods one by one:

• Weeks 1 to 4: PCA and kernel PCA, K-means clustering, estimation and Gaussian mixture models.
• Weeks 5 and 6: least squares regression, ridge and LASSO.
• Weeks 7 to 9: K-nearest neighbours, decision trees, naive Bayes, perceptron, logistic regression.
• Weeks 10 to 12: support vector machines, bagging and boosting, neural networks.

For each method, the course page says you should know the model, the loss function, the optimisation method and how it is evaluated.

What MLP covers

MLP is the practical partner. It uses scikit-learn for an end-to-end project workflow, regression with gradient descent, regularised models, binary and multiclass classification, support vector machines, decision trees and random forests, neural networks and unsupervised learning.

The MLP project runs on Kaggle for 12 to 15 weeks, split into milestones. The handbook says your L1 viva is scheduled once you cross the Kaggle cut-off and declare that all models are done. The course page says an industry expert conducts the viva. How viva cut-offs turn into grades is in how projects are graded.

How to prepare for each course

For MLF: work the maths by hand

Take tiny data. Fit a line through the origin, y = wx, to the points (1, 2), (2, 4) and (3, 5). Least squares gives w as the sum of x times y divided by the sum of x squared: (2 + 8 + 15) / (1 + 4 + 9) = 25/14, about 1.79. Doing this once by hand makes the matrix version easier to follow.

For MLT: one card per method

Make a single page for each method with four lines: the model, the loss, how it is trained, and when it overfits or underfits. Before quizzes, test yourself by covering three lines and recalling them from the fourth.

For MLP: build the habit of a clean workflow

• Split your data into training and test sets first, and do not look at the test set while tuning.
• Use cross-validation to compare models.
• Keep a notebook log of what you tried and the score it got.

For the project: start early

Start the Kaggle milestones in week 1. Your model must cross the cut-off, but it does not need to be fancy. It needs to be yours, and you need to explain every step in the viva. Copying a notebook is plagiarism. The handbook says a plagiarised project gets a U grade, may lead to disciplinary action, and must be redone at the full project fee. Graded assignments are covered by the honour code too, so never share or copy answers.

How much time to plan

The DS academics page expects about 15 hours a week per diploma course. MLF and MLT together is roughly 30 hours a week. Plan the MLP project on top of MLP. Remember that your diploma project CGPA must be 7 or more to enter the degree level.

Common questions

Should I take MLF and MLT in the same term?

You may, since MLF is a corequisite of MLT. If Maths 2 was hard for you, take MLF first so MLT does not pile up on shaky maths. See prerequisites explained.

Do I need Stats 2 for these courses?

Stats 2 is a foundation course, so you will have passed it before any diploma course. Its estimation topics return in MLF weeks 10 to 12 and MLT weeks 1 to 4. Our Statistics guide helps you revise.

Is the Diploma in Data Science a better first choice than Programming?

It depends on your goals and strengths. If you enjoyed Maths 2 and Stats 2, the ML courses may suit you as a start. Our post on which diploma to start first compares both options.

Is MLP only about coding?

Mostly, yes, but the methods come from MLT. If you understand why a model overfits, you will know what to change in scikit-learn.