All language subtitles for 04 - Features and Labels.en

[Autogenerated] before we talk about

future engineering, let's talk about what

features and labels mean when you're

building and training your machine

learning. Marty, if you're taking this

course, you've been introduced. The

concept of machine landing a machine

learning algorithm is an algorithm that is

able to learn from data. It's able to look

at data and find patterns and use these

patterns to make analysis. Machine

learning models have the ability to work

with a huge base of data and find patterns

within this data. These are patterns that

are not easily discoverable using, say,

exploratory data analysis or

visualizations. One thing's patterns have

been determined. Machine learning

algorithms can then be used for

prediction. Machine learning models, once

it's learned from your data, are capable

of making intelligent decisions. No

machine learning is a vast field with many

applications, but at its core, machine

learning problems can be divided into four

broad categories. The first is

classifications user model, tow, classify

instances, good or bad girl or boy cat or

dog classification. Models are used to

predict classes or categories. If you want

to predict a continuous numeric value, you

lose a regression model regression

analysis of what you'll do to predict,

say, the mileage of an automobile, the

price off a stock or a home. If you have a

large corpus off data and you want to find

logically, groupings are patterns that

exist in your data. You can apply a

clustering model Clustering model tries to

bring together those data points in a

single cluster, which are similar to one

another. And finally, if the entities in

your data have many characteristics are

features and you want to find which

features are important, extract latent

features from your data. You'll apply

dimensionality reduction as a student of

machine learning. These are the 1st 4

broad categories of machine learning

techniques that you'll encounter. Let's

understand what machine learning is by

taking an example off classifications. You

want to determine whether wheels are fish

are mammals. Now you know that wheels are

members of the infra order, said Asia,

which indicates that they're mammals. But

on the other hand, if you look at the

character, the sticks off a real they look

like fish. They swim like fish. They move

it fish, they live it fish in the sea.

They could be fish Now for your classifier

you essentially want to be able to feed in

the character the six off a veil into your

machine learning markets. You have an ML

based classifier, which has been trained

on a huge corpus off data. Once it has

been trained, you'll feed in the character

the sticks off a well and hope for a

prediction that is correct. If it's a

robust, welcoming classifier it he

correctly identify the wheel as a mammal.

Now the question is, how did we get this

ML base classifier trained on a corpus off

data. Then you start off with a machine

learning algorithm and you feed in a huge

corpus off training data to train your

machine learning algorithm. This

classification algorithm will look through

all of the samples present in the training

data and try toe extract significant

patterns, and this, in turn will give you

an ML base classifier. A classifier is a

fully trained model. The algorithm learns

from data to give you a trained morning.

That is your classifier. If you have a

good model, you should be able to give it

information and have it make predictions

the characteristics off well that you feed

into your machine learning model, whether

for training or for prediction, is a refer

to as the feature vector. These are the

features off your instance at the other

end the output off your model. The

prediction that it makes whether it's an

output category or a continuous of values

such as a stock price, is referred to as

the label. Now it's quite possible that

you feed into your machine learning model

entirely different characteristics off.

Will you tell it that it moves like a fish

and it looks like a fish Now, in such

situations, your classifier is lightly toe

indicate that the wheel is a fish, which

is clearly wrong. What you fed in in your

input feature vector are incorrectly

specified features, and when the features

that you feed into your model are not set

up correctly, you're likely to get an

incorrect prediction from your model. Your

model is only as good as the features that

you use for training the features that you

used to train your model or also refer to

US ex variables. X variables are the

attributes that the machine learning

algorithm focuses on the characteristics

of the entities on which your training

your model of the attributes are ex

variables, and every data point is a list

or vector off such ex variables, and this

is what is together refer to as a feature

vector does The important animal algorithm

is a feature vector feature vector exactly

the same thing as ex variables, and I'll

use the storm's interchangeably throughout

the scores. The output of your machine

learning model, the predictions that it

makes I refer to us. Why variables the

attributes that the machine learning

algorithm tries to predict our called

labels, or by valuables once again, a

loose the terms labels and by variables

interchangeably, they mean the same thing.

Another time that you can use for by

variables are targets. Now, based on the

kind of model that you're building, the

labels can be off different types. If you

have categorical or discreet label values,

that is typically the output off a

classification algorithm. Spam or ham,

True or false? A, B, C or D. These are

examples of categorically values or

labels. Labels can also be numeric or

continuous values. These air typically the

output off what regression models such as

the models that you use for price

prediction. When you're working with

continuous output from your machine

learning Marley, you might call them by

values or targets rather than labels,

because labels have a very categorically

field. Now that you understand the

difference between features and labels,

there is an important point that you need

to remember garbage in garbage out. If the

data that you feed into an ML model is off

a poor quality, the model itself will be a

poor model. Your mortal is only as good as her data.