Dimensionality Reduction

Loading...

From the course by IBM

Advanced Machine Learning and Signal Processing

8 оценки

IBM

Advanced Machine Learning and Signal Processing

8 оценки

Course 2 of 4 in the Specialization Advanced Data Science with IBM

>>> By enrolling in this course you agree to the End User License Agreement as set out in the FAQ. Once enrolled you can access the license in the Resources area <<< This course, Advanced Machine Learning and Signal Processing, is part of the IBM Advanced Data Science Specialization which IBM is currently creating and gives you easy access to the invaluable insights into Supervised and Unsupervised Machine Learning Models used by experts in many field relevant disciplines. We’ll learn about the fundamentals of Linear Algebra to understand how machine learning modes work. Then we introduce the most popular Machine Learning Frameworks for python Scikit-Learn and SparkML. SparkML is making up the greatest portion of this course since scalability is key to address performance bottlenecks. We learn how to tune the models in parallel by evaluating hundreds of different parameter-combinations in parallel. We’ll continuously use a real-life example from IoT (Internet of Things), for exemplifying the different algorithms. For passing the course you are even required to create your own vibration sensor data using the accelerometer sensors in your smartphone. So you are actually working on a self-created, real dataset throughout the course. If you choose to take this course and earn the Coursera course certificate, you will also earn an IBM digital badge. To find out more about IBM digital badges follow the link ibm.biz/badging.

From the lesson

Unsupervised Machine Learning

Curse of Dimensionality9:25

Dimensionality Reduction4:02

Principal Component Analysis6:02

Principal Component Analysis (demo)6:51

Covariance matrix and direction of greatest variance8:21

Eigenvectors and eigenvalues8:33

Projecting the data4:29

PCA in SystemML2:34

Meet the Instructors

Romeo Kienzler
Chief Data Scientist, Course Lead
IBM Watson IoT
Nikolay Manchev
Data Scientist
IBM EMEA Data Science

In order to combat the class dimensionality,

people use dimensionality reduction.

And again this is an umbrella term for various techniques.

For example, feature selection.

And the idea of feature selection is,

if you have a dataset with many many features,

you probably don't have to train your classifier using all of them, instead,

what you can do is you can hand pick specific features that have

special predictive power to give you a good prediction for the target that you build,

instead of blindly using all features that are available.

Feature selection can be automated.

People use things like information gained from information theory,

like similarly to how a decision through classifier's work.

People also use domain knowledge.

So you might use experts that can tell you which

attribute has sufficient predictive power for the problem you're trying to solve.

For example, let's say that you have a dataset with a different view course,

so it's an ultimate multi-dataset and you have data for different cars

or specific models and manufacturers and you have data like, I don't know,

you have data like the weight of the vehicle,

you have data like the color,

you have data like the brake horsepower,

you have data about the maximum RPM's,

you have data about the fuel type, and so on.

And let's say that you want to predict the MPG, the miles per a gallon.

So, you could, in theory,

use all these attributes and try to train the classifier,

but it will be better if you actually speak with

an automotive engineer and they will probably tell you

that the mpg is mostly influenced by maybe the weight of the vehicle,

and I don't know, the fuel type,

but it's definitely not influenced by,

let's say the color of the vehicle,

so you can remove this attribute from your input dataset when training the classifier.

Another approach of tackling the class of dimensionality is to do feature reduction and

the idea of feature reduction is that you have a number of features,

let's say you have features like x_1,

x_2, x_3, x_4, x_5,

and you feed all these guys to

something and then at the end you end up with a new set of features,

maybe just x_1 and x_2,

but these features retain most of the information that was already delivered by x_1,

x_2, x_3, x_4, and x_5.

So, the idea of feature reduction is use all the features from the dataset, right,

and then generate a dataset with small number of

dimensions while at the same time

retaining as much information as possible from the original features.

And there are different techniques for feature reduction.

Things like principle component analysis, linear discriminant analysis,

here we have multi-dimensional scaling,

but there're also non-linear methods like self organizing maps,

autoencoders, and so on and so on.

But in this session,

we will focus on principal component analysis,

as this is a really universal method.

It's widely used and it is extremely useful.

So, in the next video,

we'll talk about PCA.

Ознакомьтесь с нашим каталогом

Присоединяйтесь бесплатно и получайте персонализированные рекомендации, обновления и предложения.

Coursera делает лучшее в мире образование доступным каждому, предлагая онлайн-курсы от ведущих университетов и организаций.

© Coursera Inc., 2018 Все права защищены.

Загрузить из App Store

Загрузить в Google Play