Retour à Mathematics for Machine Learning: PCA

4.0

1,048 notes

•

217 avis

This intermediate-level course introduces the mathematical foundations to derive Principal Component Analysis (PCA), a fundamental dimensionality reduction technique. We'll cover some basic statistics of data sets, such as mean values and variances, we'll compute distances and angles between vectors using inner products and derive orthogonal projections of data onto lower-dimensional subspaces. Using all these tools, we'll then derive PCA as a method that minimizes the average squared reconstruction error between data points and their reconstruction.
At the end of this course, you'll be familiar with important mathematical concepts and you can implement PCA all by yourself. If you’re struggling, you'll find a set of jupyter notebooks that will allow you to explore properties of the techniques and walk you through what you need to do to get on track. If you are already an expert, this course may refresh some of your knowledge.
The lectures, examples and exercises require:
1. Some ability of abstract thinking
2. Good background in linear algebra (e.g., matrix and vector algebra, linear independence, basis)
3. Basic background in multivariate calculus (e.g., partial derivatives, basic optimization)
4. Basic knowledge in python programming and numpy
Disclaimer: This course is substantially more abstract and requires more programming than the other two courses of the specialization. However, this type of abstract thinking, algebraic manipulation and programming is necessary if you want to understand and develop machine learning algorithms....

Jul 17, 2018

This is one hell of an inspiring course that demystified the difficult concepts and math behind PCA. Excellent instructors in imparting the these knowledge with easy-to-understand illustrations.

May 01, 2018

This course was definitely a bit more complex, not so much in assignments but in the core concepts handled, than the others in the specialisation. Overall, it was fun to do this course!

Filtrer par :

par Chuwei L

•Apr 05, 2019

worse than previous courses of machine learning specialization. Really confused me when introduced the inner products.

par gaurav k

•Jul 03, 2019

More examples and visualization should be there to explain.

par Mark P

•Jul 30, 2019

This course had a lot of potential but there were a number of inconsistencies, cut/paste comment bugs, that make it more challenging than it needs to be. The comments in the notebook exercises should be triple-checked with the text above to ensure consistency of variables. Far too often these would be mixed up, or the input/output descriptions would be incorrect. Or the unit test would have different dimensions. Lectures often left out steps - e.g. "because of orthonormal basis, we can simplify and remove a bunch of terms" - how exactly? A extra few seconds of explanations would allow students to follow more closely. Notation in lectures is sloppy - sometimes terms would be missing and then the video would quietly cut to a correction. "j's" and "i's" indices were interchanged frequently making the derivations how to follow. Also, this isn't a course on unit testing - some more tests should be included to help students debug individual functions rather than relying on the final algorithm (e.g. PCA to work). It should be explained why the "1/N" term for XX^T is not necessary even though it's in the lectures. On the plus side, the added written notes were welcome and fairly well done.

par Omoloro O

•Aug 07, 2019

Compared to the first two courses in this specialisation, this course was not very engaging. Additionally it was often hard to see what the end-goal was and the instructor seemed to be going deep into details without making the practical reasoning behind it clear. Furthermore, a lot of the exercises involved repetitions of tasks that can easily be done by computers.

par Ben H

•Aug 20, 2019

This course had a lot of potential, but unfortunately the pacing, structure, and teaching was not up to the standard of the other two courses in the specialisation. The teacher is clearly very knowledgable about his subject, and seems like a really nice person, but delivers the material in a very direct, formal mathematical style. This makes it much more difficult to gain intuitive insight into the subject matter.

Given the level of the past two courses, this felt like way too big a leap. Don't get me wrong — this course is still worthwhile, but could use some refining.

par Hsueh-han W

•Sep 20, 2019

many steps are not clear enough that I have to spend a lot of additional time to figure out the details.

par Adrian C

•Sep 22, 2019

The derivatiion of the PCA in the last week can be broken into 2 weeks with different programming assignments to get a closer and more confident understanding of the PCA method.

par Gurudu S R

•Sep 16, 2019

Tutor is not clear and concise on the concepts. Need more examples for Week 2 and Week 3.

par Jean D D S

•Aug 31, 2019

I would ask the lecturer to go on more detail on the explanations and do (more) examples.

The lecturer tends to skip a few steps during calculations and demonstrations.

par Rafael C

•Sep 24, 2019

The Classes didn't give the knowledge to solve the assignments.

par Nont N

•Sep 25, 2019

I am a bit disappointed by this course. The professor didn't do much to help learner understand what's the meaning of the math we are looking at. Much of the quiz is just math grinding. The programming assignment require a lot of my effort in programming, but not much on math.

I'm not saying that this course is very bad, but Compare to the previous 2 course in the Math for ML specialization, provided by the same university, this one is obviously inferior.

par Marina P

•Sep 06, 2019

The course is interesting, but some of the quizzes were not done very well. After the first 2 parts of this course, which were just amazing, this one seems kind of worse, although by itself its not that bad.

par amit s

•Feb 08, 2019

Unlike the prior courses in the series, topics not clearly explained and brought too sudden. Also none of calculations shown completely, instructor just wrote results in the end. Due to all these reason I was not able to finish the course.

par Mark C

•Jul 31, 2018

Only on week 1 but this is already a disappointment compared to the first two classes in the Math for ML series which were excellent. Some content is presented too fast. Quiz questions are ambiguous. I already paid for the class so I will finish the content but not worry about passing quizzes and assignments. Had I known it would be like this I wouldn't have paid for it. Check out the other reviews and forum discussions to see what others think.

par ABHI G

•Aug 22, 2018

not so good

:(

par Nouran G

•Oct 11, 2018

Course is inconsiderate to new learners in that new concepts were very sloppily introduced. Like the first two courses of the specialization, this course is shallow, shouldn't be anyone's introduction to the subject and is a refresher at best. Unlike the other two courses, it assumes python knowledge, doesn't explain relevant syntax in the assignments; which made me take a lot of long unnecessary detours to get the python implementation right.

par Vignesh N M

•Sep 12, 2018

Explaination of many things are skipped, assumption was made by the instructor that lot of things were already known by the learner. It could have been much better.

par Kevin L

•Sep 11, 2018

The course assignments could be improved dramatically, though the course itself has very good content if you want to have a taste of how linear algebra (predominantly) can be implemented to solve machine learning problems.

par Daniel U

•Sep 27, 2018

Programming assignments seemed to be written from a completely different direction, and instructions are vague and misleading. (The math assignments were not so bad.) There was no staff or mrntor engagement in the forums during the period of the course.

par Marvin P

•Apr 24, 2018

After the other two awesome courses of the specialization this one stays far behind my expectations. Weakest course of the specialization. Instructor is obviously knowledgeable but does not provide much intuition. Programming assignments are really difficult and at many points frustrating. 2 more weeks and therefore comprehensive instructions would be desirable. Couldn't appreciate that course as much as I wanted to.

par Scoodood

•Jul 28, 2018

Video lecture not as intuitive as previous courses.

par Xiao L

•Jun 03, 2019

very wired assignment, a lot of error in template code. The concept is not clear.

par Tobias T

•Jul 14, 2019

If you like traditional lectures, which you go into a math class then feel puzzled, then go for it. Otherwise, the contents of this course are simply going through the mathematics equations and definitions, which can easily be found in textbooks. Ironically, the previous two courses in this specialization used lots of graphics and animations to help you understand the maths (either in terms of equation-wise or intuitively), this course completely lacks this element.

par Michael D

•Jul 22, 2019

After having done the first two parts of the specialization, I am afraid this one didn't stand up to the high quality bar the previous two had set. The programming assignments are unnecessarily long and complex and the overall material is not as engaging, connected and concise. I might give it a good rating as a standalone but now I can't avoid comparing it to the other two parts of the specialization.

par Max B

•Aug 14, 2019

Pretty bad all around.

The teacher keeps throwing formulas without taking the time to explain why they are useful, and what they represent.

The first two courses were really good, and this one is a bummer.

Most of what I learned was learned elsewhere, the course acted as a detailed syllabus with some practice quiz (of relatively poor quality).

It's still worth taking if you completed the first two courses and want the specialization certification.