Lecture 12: Data Modelling for Machine Learning

Nafees AI Lab

Data modeling is the process of creating a blueprint or visual representation of how data is organized, structured, and related within a system. It helps to understand and define the data types, their attributes, relationships, and how they are stored and used. This process is crucial for designing efficient and effective databases and information systems.

Transcript

00:00Hello guys, welcome back.

00:05So, now we have our fifth feature, which is basically modeling.

00:10Modeling is what is called?

00:11Modeling is three parts.

00:14The first part is that you have to choose and train.

00:20Model choose and train.

00:23Train and choose model.

00:25Part one is called.

00:26Part two is that you have to find.

00:30Tune the model.

00:36And third part is that you have to compare.

00:41Model compare.

00:44Comparison or compare.

00:46These three things you have to be in mind.

00:49Modeling is three basic components.

00:51Modeling is first.

00:53Model choose and tune.

00:54Then you have to compare.

00:56Modeling is first.

00:57The same thing you have to choose.

01:00If you have turned to a machine learning learning page.

01:02To be patient, that is the basic unit.

01:05If you have three tasks.

01:06It doesn't matter.

01:07To be patient, of course.

01:08You have to try to train.

01:09The same thing that you have to do.

01:10If you have three tasks as a team,

01:10So if you have three tasks,

01:13if you have three tasks for three tasks,

01:15whether you have model training or tune or model compare,

01:20you have to do data.

01:21So this means that you have to do data set.

01:28And remember that the data set is different,

01:32one data set is same.

01:34So if you have to say 100 patients,

01:39if you have to say 70 patients,

01:43you have to do data and then you have to do data.

01:48Then I will tell you that 15% of the data you have to do data.

01:55And then 15% of the model test.

02:00Make sense?

02:02Three stages are.

02:04Train, tune and model comparison.

02:07When you train, 70% of the data you have to do model training.

02:13Toon, 15% normal practice.

02:16But normal 70, 15% of the data you have to do.

02:20Now I will repeat it so that you have to do confusion.

02:26So this example you have to do a course that you have to do.

02:30prepare you university in the middle of the semester.

02:32Now normal is that you have to study

02:34courses and study courses.

02:36Then when exam comes,

02:38you have to train

02:40all the semester.

02:42You have to train

02:44all the semester.

02:46Then exam

02:48you have to test exam.

02:52You have to practice exam.

02:54This question, exercise

02:56that I have done.

02:58You have to validate

03:00you.

03:02Validation data set

03:04is called

03:06validation

03:08data set

03:10and finally

03:12you have to do

03:14actual test data.

03:18The model comparison

03:20is called test data

03:22and the tuning model

03:24is called validation data

03:26which basically

03:28train

03:30is called train data.

03:32Make sense?

03:34those three sets

03:36are.

03:38You have to choose

03:40and train

03:42and train

03:44you have to do it.

03:46You have to do it.

03:48You have to do it.

03:50You have to do it.

03:52You have to do it.

03:54You have to do it.

03:56Basically

03:58and finally

04:00I have to do it.

04:02I have to do it.

04:04You have to do it.

04:06You have to do it.

04:08You have to do it.

04:10So if you have to do it,

04:12you have to do it.

04:14If you have a test data

04:16you have to train

04:18if you have to do it.

04:20You can't show it.

04:22So this is why this is important for the accuracy of the model.

04:33This is why this is important for the three data sets, whether it is training or validation or model,

04:41it will be separate to the model.

04:46The example I have given is that in machine learning,

04:56if you have the test data to train,

05:01the accuracy is 100%, marks are 100 by 100,

05:06you can represent the true capability, not.

05:11This is called overfitting.

05:13In machine learning, this is called overfitting.

05:18On the other hand, if you have the test data,

05:21if you have the test data,

05:23this is called overfitting.

05:25This is the terminology,

05:27which is called underfitting,

05:31it is called underfitting.

05:33Okay?

05:34Underfitting.

05:36Okay?

05:37This is the two things you have seen.

05:40Basically,

05:41in machine learning terms,

05:44if you have the test data to train,

05:50the accuracy is 100%,

05:52marks are 100 by 100,

05:54but the markers are the true capability,

05:57you can represent the true capability.

05:59It is not.

06:00It is called overfitting.

06:03In machine learning terms,

06:05it is called overfitting.

06:08On the other hand,

06:09you can't do it.

06:10The test data is called overfitting.

06:12You can't do it.

06:13This terminology,

06:15which is the accuracy,

06:18the accuracy is low.

06:20It is called underfitting.

06:22Underfitting.

06:23Okay?

06:24Underfitting.

06:25Underfitting.

06:26Okay?

06:27Okay.

06:28These two things you have seen.

06:29Basically,

06:30there are three basic components.

06:33I repeat and then,

06:34I repeat and then,

06:35then,

06:36then,

06:37you can choose and train.

06:39But,

06:40when I talk about it,

06:42when I talk about it,

06:43in the background,

06:44on the context of it,

06:45there is data.

06:46There is data.

06:47You can train,

06:48tune,

06:49and compare it.

06:50You can compare it.

06:51What is the model?

06:52You can compare it.

06:53I am sure.

06:54I don't know,

06:55so.

06:56In the case of,

06:57this data,

06:58I know,

06:59the data,

07:00the data,

07:01which we have used,

07:02the data.

07:03As we say,

07:05the data,

07:06is the data.

07:07It's got a trained data.

07:09Which we have used,

07:11Now,

07:12which we have used,

07:13which we have used do,

07:14which we do.

07:15The model true.

07:16We have used to know

07:17how to change these items.

07:18it is like this example, if you have a practice exam, practice exam is so important,

07:23that you can do it, you can train, that you can validate it,

07:28then the exam on the validation, which we call the machine learning, validation data.

07:36And the final data, which you have final test, which you call test data,

07:42and then you compare the model.

07:45Now, the model tuning is very important.

07:49In comparing, we will go to the model tuning which is very important.

07:53So, I will do this in this way.

07:56So, the model tuning is on the model tuning.

08:00You can see that the model accuracy is not good.

08:04So, in every model, whether you go to the regression model,

08:10or in the neural nets, or in the random forest,

08:16there is one thing that exists.

08:20It is called hyperparameter.

08:26Hyperparameter.

08:28Hyperparameter, you can understand that

08:31it is a small button that you can change your car.

08:38You can imagine that in the cockpit,

08:41there are many buttons that you can change your car.

08:44Then, you can change your car.

08:47In the machine learning,

08:50there are hyperparameter.

08:52So, you can play the hyperparameter.

08:54So, you can play the model of fine tuning,

08:58or the result,

08:59the required result will be achieved.

09:02So, for example,

09:04you can play the other part.

09:06Like this, you can play the DJ.

09:08You can play the DJ,

09:09or you can play the music,

09:10or you can play the music.

09:11It will be fine tune for experience.

09:13Okay.

09:14Hyperparameter, you can also have idea.

09:16Excellent.

09:17So, you can do it here.

09:19Let's close.

09:20Now, we have only one thing left.

09:22Validation, we have also learned about the data splitting.

09:25We have learned about the data splitting.

09:273-4 concepts

09:29basically

09:31model validation

09:33next lecture

09:35we will see

09:37modeling

09:39basic parts

09:41split

09:433 parts

09:453 different data types

09:47training validation

09:49test

09:51tuning

09:53tuning

09:55hyper parameter change

09:57training

09:58basically

09:59we have seen

10:00model

10:01train

10:02this

10:04model

10:05accuracy

10:06and

10:07compromise

10:08time

10:10accuracy

10:11model

10:1290%

10:13accuracy

10:14computer

10:15run

10:16so

10:1790%

10:18customer

10:19run

10:20so

10:21you

10:22train

10:23accuracy

10:24so

10:25this

10:26training

10:27time

10:28versus

10:29computer

10:30accuracy

10:31a

10:32c

10:33versus

10:34a

10:35c

10:36u r a c y

10:37accuracy

10:38so

10:39these

10:40two

10:41things

10:42that

10:43time

10:44versus

10:45accuracy

10:46time

10:47accuracy

10:48we can

10:49discuss

10:50we can

10:52discuss

10:53this

10:54lecture

10:55close

10:56next

10:57lecture

10:58model

10:59comparison

Category

Transcript

Recommended