0:05

Okay, the good news.

This course is very practical.

There's very little maths involved.

But, you knew there was but, didn't you?

I do expect you to know the difference between mean, median, mode.

Mean is the arithmetic average.

Median is the middle number in a sequence.

Mode is the most common number.

I expect you to know what a normal distribution is, also called Gaussian.

And I expect you to know what a standard deviation is,

and how that relates to the normal distribution of course.

I would also expect you to be familiar with say, exponential probability distribution.

This might represent the time between independent events as well as perhaps,

the Poisson distribution, which is

representing the number of independent events per time period.

Though the thing about this,

if the events are frequent,

say more than 20 or 30 per time period,

then it can just as well be modelled by a normal distribution.

1:51

Okay, I hope that was easy.

Programming, I expect you to be able to program.

This course shows all examples in R and Python.

So, ideally, you need to be familiar with either R or Python.

However, if you are familiar and comfortable with another programming language,

you should be able to get through.

You'll have to do a bit of web searching to work

out syntax differences but you'll probably get through.

However, if you've never programmed before,

and don't know any programming language,

I think you should go away and learn a programming language,

do another course, learning say,

Python or R, and then come back and do this course.

Because you do need to be able to write scripts.

Okay, the next topic,

I'd like you to be already familiar with,

is the difference between bias and variance.

But in case you're not, we're just going to do it very quick with you.

So, imagine we're shooting at this target.

Bias is how far away you are on average from the correct answer,

from the center of the target.

Variance is how wild a shot you are.

How far away you're answers on average tend to be from correct.

So this first example is a high bias and a high variance.

We're not getting anywhere near the center of the target,

and they're all over the place,

like a drunken shotgun.

The next example shows some bias, but low variance.

All points are clustered in about the same part of the target.

This one is low bias but high variance.

So they are all over the place but on average,

they were around the best answer.

And finally, the ideal,

what we want every model to be giving us is low bias,

low variance, and they all clustered around the center of the target.

Sometimes you're going to be doing a tradeoff between

bias and variance when making models,

so it's important to understand those two concepts.

The last thing I want to talk about is confusion matrices.

This is a simple two by two confusion matrix.

So the numbers represent the number of true positives,

false positives, true negatives, false negatives.

And the ideal is to get the numbers on the diagonal high, and the other numbers low.

So you want high true positives and high true negatives.

You often see it as a two by two grid,

but when we're doing multinomial classifications,

it can be a bigger grid. This one is 10 by 10.

And again, the ideal is to have high numbers on the diagonal.

Okay, after this video is a text file with suggested reading for each of these topics,

do make a point of skimming through.

Because if students have given us feedback of

other things they wish they'd known before they started the course,

we'll add them into that file.