ETC5250 Introduction to Machine Learning


4 Star.PNG

Difficulty:

Year Completed: Semester 1, 2021

Prerequisite: ETC2420 / 5242

 

Exemption:

Actuary Program - Core Data and Statistical Analysis:

ETC5250 (100%)

Minimum of 70 for ETC5250 plus satisfactory completion of actuarial project for Data Analytics Principles.


Mean Setu Score: 75.97%

 

Clarity of Learning Outcomes: 80.00%

Clarity of Assessments: 73.60%

Feedback: 72.60%

Resources: 72.20%

Engagement: 85.20%

Satisfaction: 72.20%


Subject Content:

Lecture(s) and Tutorial(s):

Lecture Recording:

Textbook(s):

Assessments:

 

Week 1: types of machine learning problems, bias-variance

tradeoff

Week 2: review of regression, flexible regression techniques

(splines, GLMs)

Week 3: logistic regression, resampling (bootstrap, cross

validation)

Week 4: dimension reduction: linear discriminant analysis,

principal component analysis

Week 5: visualising high dimensions: tours and parallel coordinate

plots

Week 6: classification and regression trees

Week 7: random forests, support vector machines

Week 8: introduction to neural networks, regularisation

Week 9: tools for model assessment

Week 10-11: clustering 

Week 12: discussion of Kaggle project

2 x 1 hour lecture

For students seeking exemption from the Institute subject there are three additional 1-hour classes to attend throughout semester (not weekly)

1 x 1.5 hour lab

Just going through tutorial questions and working through issues

as a class. The labs are based mostly in R with application-style

questions, but occasionally the textbook is referred to for

questions testing theory. Generally we are given time at the start

of the lab to work on the questions in small groups, then about

halfway through we discuss as a class.

The textbook Introduction to Statistical Learning (ISLR) is used

extensively and is very helpful. Its explanations are very clear and

easy to follow, and fills in many of the gaps in the lecture notes.

Most of the lectures are based directly off this textbook. The

textbook is freely available to download.

Two assignments (10% each)

Project - Kaggle competition (15%)

Weekly quiz x 11 (5% total)

Exam (60%)

For students seeking exemption from the Institute subject there is

an additional assignment worth 10%. The other assessments are

all re-weighted to be worth a total of 90%.

All assignments were individual. The weekly quizzes are very easy

to complete, but the assignments and the project all required

quite a lot of work.



Comments

The unit covered many interesting concepts, some in depth but

some at a surface level only. It provides a useful starting point for

further learning in the area.

Lectures were quite engaging (online only this semester) and were

evenly split between theory and examples. Unlike similar units

(ETC3550/5550, ETC3580/5580), there's almost no coding done in

the lectures, although the Rmarkdown file for each week's

lectures contains the code for all the examples that are in the

slides. The slides are not very detailed so it helps to hear Di's

explanations and read the relevant sections in the textbook for

more detail.

 

Lab attendance is not mandatory. Occasionally there was some

material covered in labs that wasn't covered in great detail in the

lecture. The solutions are provided before the class. Preparation

isn't necessary, but since the solutions are provided I found it

helpful to work through the questions beforehand so that I had

more class time to ask questions.

Quizzes were very easy and could usually be completed by

skimming through the lecture notes.

The two assignments are based in R, and involve a mix of

theory/application questions. They were quite lengthy and

challenging, but the marking scheme is clear and detailed

feedback is provided. They were very helpful in clarifying my

understanding of certain concepts. 

The project involves building a model to make predictions on a

given dataset. It is conducted via Kaggle and part of the mark is

based on the accuracy of your predictions. Unlike previous years,

the mark is not based on your rank within the class.

This year's exam was a 70-something question exam, mostly

multiple choice with a handful of short-answer questions. A

practice exam of a similar format (but much shorter length) was

provided. Most questions only tested surface-level understanding,

unlike the assignments. Negative marking was applied to the

incorrect multiple choice answers. This was not communicated to

students before the exam. The quizzes and the exam both

suffered from typos and poor wording that caused confusion for

some students so it is advised for students to be careful. The

assignments were generally much more refined.

Don't underestimate the difficulty of the course. Some topics may

seem simple but the quizzes and exam standard required a solid

understanding of the details behind it. Read the textbook, don't

rely on lectures alone.

General Overview:

Lectures:

Tutorials:

Assessments/Other Assessments

Exam

Concluding Remarks