Original subtitles

Welcome back.

Up until now I've been blabbering on about what machine learning is and I hope you have a bit of an

idea of why we have it and why it's useful.

But at the end of the day the only reason that we care about machine learning is that we're able to

use machines to predict results based on incoming data.

That's it.

Now this idea of machine learning.

And don't worry.

I know we keep talking about theory but I promise we're gonna get some coding exercises first.

But I do want to talk quickly about some of the machine learning categories that you often see and keep

in mind what I said that is machine learning is simply about predicting results based on incoming data.

And all these subcategories simply do that.

For example we have the idea of a supervised learning which is a subset of machine learning in this

supervised learning.

The data that we received already has CATEGORIES.

THINK OF IT AS A CSC files with rows and columns label.

We have labeled data and a test data that is label so we know if our function is right or wrong.

So in a supervised learning scenario we can do things like classification to decide is this an apple.

Or is this a paper machine learning model simply draws a line to decide Hey this is an apple and this

is a pair or we might do something called regression instead of classification based on inputs.

For example predicting stock prices another way that we might use supervised learning is for example

to hire engineers based on inputs based on years of experience based on maybe age maybe where they live

what type of computers they have all these sorts of inputs that are labeled can be used in a supervised

learning system to decide should I hire this engineer or should I not.

Now sometimes we have data that doesn't have labels and this is called on supervised learning.

Again think of it as a CSP file without perhaps the column names labeled sometimes with things like

clustering.

We need to create these groups or at least the machine to create these groups.

For example we give it a bunch of data points and then the machine decides oh this is a group.

This is a group and this is a group or we can have something like association rule learning where we

associate different things to predict what a customer perhaps might buy in the future when groups don't

exist.

We call it unsupervised learning.

We can't tell the machine that they are right or wrong like we can when we do apples versus pears since

there are no true categories but we let the machines just create these categories for us.

Finally we have this idea of reinforcement learning and reinforcement learning is really interesting

because it's all about teaching machines through trial and error through rewards and punishment so the

program simply learns a game by playing it millions of times until well it gets the highest score it

doesn't know what it's doing at first but then it tries to maximize the score and eventually figures

out that hey maybe I should try and move where the ball is coming.

So this is seen for skill acquisition or real time learning and you see it a lot in machine learning

programs that play for example video games.

But the idea here is that machine learning has different categories and different ways to accomplish

its goal and topics like neural networks decision trees support vector machines K nearest neighbor are

simply algorithms that are used with these sub fields in order to come to these predictions but remember

the key thing all of these what they're doing is trying to learn from the data that it receives and

predict something I'll see in the next one by.