Problem Set 1

The two parts that happen here

Problems 1 and 5 of PS1 are on this site. Everything is recorded as you go, so there is nothing to zip and nothing to upload. The rest of the problem set is on intro2ml.com.

Problem 1: Warm Up

20 points. Ten claims about linear regression. Some are true.

Answer here. Pick true or false, and say why in a sentence or two. The box has to be filled before the answer will submit, which is on purpose: the reasoning is what is being read.
The reason carries the marks. The verdict is one of two options and you can guess it. What you write about why is the part that shows whether you have the idea.
Do it after the lecture and the reading, not before. The point is to find out what actually stuck.

Argue with it. Every question has a thread underneath. Before you answer, the tutor will not tell you which option is right, and it will say so if you ask. Everything else is open: definitions, a worked example, the mathematics, or a question back at you.

A note on getting it wrong Several of these are false, and the false ones are the interesting ones. Being wrong here costs nothing and tells you exactly where to look.

✍ Check Your Understanding

That was Problem 1

Your answers and your reasons are saved as you go. You can come back and change your mind: the later answer is the one that counts, and what you wrote the first time is still there.

Problem 5 is next, and it is a longer piece of work. Give it a proper sitting rather than squeezing it into ten minutes.

Problem 5: Learning ML with ML

20 points. One sustained conversation about linear regression.

What you have to do in it

Answer the first five questions it gives you
Ask for five more, harder ones. Ten in total
Ask for at least one derivation
Ask for at least one piece of code with a plot
Ask two questions of your own that go past the lecture
Ask for a summary and two next steps
Write the reflection, in the conversation

How it is marked, out of 20

5 conversation quality
5 the ten questions. Only 1 of the 5 is for getting them right
4 code and plots
4 reflection
2 checking the tutor

Those last 2 points used to be for handing in a tidy PDF. There is no PDF now, so they go to something the recording can actually show: at least one place where you verified a claim, asked for it to be derived again, or disagreed with a reason.

What a good conversation looks like

A conversation where you disagreed with the model twice is worth more than one three times as long where you accepted everything. Length is not the point.

Push past the first answer. The first reply is a summary. The second, after you ask for what is underneath it, is usually where the content is.
Ask for the maths when the words are vague. If a claim cannot be written down, it may not be true.
Argue back, then check. When something sounds wrong, say so and test it on a small numeric case. Models state false things fluently.
Be honest in the reflection. Write what you got wrong and why. Do not tidy the language: unedited is better here, and easier to believe.

That is both parts

Problems 2, 3 and 4 are on intro2ml.com, and go to Moodle as usual.

Found a bug or a badly worded question? Use the comment button at the top right. It goes straight to a list I read.