Original subtitles

Or Rado Rado rat modelling Part 3 churning once you're happy with your models initial performance on

your training dataset.

The next step is to try and improve it like what we said in last lesson.

How a car can be tuned for different styles of driving a model can be tuned for different types of data.

Specifically your data usually this training will take place on a validation data split.

However if you don't have access to a validation set due to how you've done your your training validation

and test data split it can also happen on the training data on many models have different hyper parameters

which can be adjusted.

You can think of these as the same as dials on an oven.

For example when you're learning to cook your favorite chicken dish that all that delicious sweet honey

mustard dish.

I'm getting hungry making these lectures.

You notice that cooking it at 180 degrees for an hour meant it came out a little raw not ideal right.

I want this chicken to be nice and crispy but at 200 degrees that's where we get the glory.

That's 200 grains for an hour.

My chicken came out just right.

So like a little dial here Ryan.

This is how you consume your oven.

You can change it from 180 grains to be 200 degrees.

If you'd lived 180 grains chicken doesn't turn out well you'd take it up 200 degrees.

Chicken does turn out well.

Depending on what kind of model you're using will depend on what kind of hyper properties you can chew.

For example a random forest will allow you to adjust the number of trees in this one we've got three

trees.

So we've got five trees and we'll have a look at the random forest.

This is a type of machine learning algorithm.

We'll check this out in the future lesson and a no network will allow you to adjust the number of layers.

Again we'll we'll look at a neural network in a future lesson just giving an example of different hyper

parameters like the temperature of an oven.

You can adjust on different kinds of algorithms.

The main things for you to remember are machine learning models have hyper parameters you can adjust

however depending on what model you're going to use the hyper parameters will be different.

The goal of tuning hybrid parameters is to improve your model's performance and you should do model

tuning on your training and or validation sets meaning you train a model using an initial set of hyper

parameters.

These might be the defaults like say for example you put your chicken ingredients in the oven you press

chicken roast chicken on the oven and it goes for an hour on the ovens default settings and then you

find it doesn't turn out how you want it to be you might adjust them as you go.

This is exactly what you do with a machine learning model.

You might try a machine learning model with an initial set of hyper parameters the default.

And it turns out it does okay but not exactly how you want it.

So you might try and improve it by adjusting some of the the hyper parameters up next we've got model

comparison.

Now this happens during your experimentation so you've trained multiple different models on the same

dataset.

We're going to look at how you might compare those and this is done on the test data.

So let's do it.