...

10.

amz

Top

Top

wow

no

ok

很好

:)

很好

ok

<3

好

-

♥

m

.

v

It's fine if you have a relatively strong background in implementing "multi-step" mathematics in Python. I would not say this is for an actual beginner. Maybe not even someone who is concurrently learning Python AND this course at the same time. I'm rating the course highly for the subject matter that it presents, but I struggled IMMENSELY during the practicals (you actually code key portions of the definitions--but not the entire Jupyter notebook).

The video portion is awesome. Andrew ("Dr. Ng"?) provides an excellent "plain-english" down-to-earth explanation of the math behind the algorithms. The code, however... Well, let's just say it FEELS like one of those art-instruction jokes: "Drawing an Owl: First you start with two circles. Second, you draw the rest of the (explicative) owl."

I don't really think I have much of a furture in anything remotely involving math and programming. After this course, I honestly feel more inclined to stick to dumping data with SQL and letting the grownups slice and dice it.

Bottom line: I've been in IT for over two decades and have alot of (outdated) skills in my toolbag, but this course brought me to tears of feeling like my brain is finally starting to slip away. I just can't learn stuff the way I used to. I don't know. Maybe this will all make more sense after I've slept on it.

Great introductory course, Andrew is really talented in making everything he says crystal clear. However, I've found few minor things I don't like:

1. As someone who started this course when it was free, I can say that previous version offered much more insights and tougher assignments and harder quiz questions, it was harder overall. This version feels kinda dumbed down a bit.

2. Some (important?) things are left unrevealed, not enough attention is paid to the issue of feature selection and feature engineering (maybe some of it will be covered to extent in other specialization courses). That last assignment that included regularization in logistic regression had already given function which mapped 2d features into 27 dimensions, and someone without much math background could not really see how to map such cases by themselves. Maybe that stuff is out of scope of this course, but whole model fails if someone doesn't know how to do that input preprocessing and knowledge about algorithm then becomes irrelevant.

That being said, whole course was amazing and interactive, with really valuable content, especially for a beginner.

4.5/5 from me

Nice course that provides an introduction to supervised machine learning and teaches you how to implement the linear and logistic regression algorithms and improve their performance. Well-explained and beginner-friendly.

Easy course, but it picks up pace towards the end. It involves graded labs in Python from the second week on, for which you need to know basic Python but they also give you plenty of hints. People who have taken calculus and linear algebra classes should have no problem following the math. I would have liked more technical and math details, but that is not the purpose of the course.

The graded labs are great for beginners, but since you only have to implement bits of the algorithms (and get lots of hints), those with scientific programming experience will not benefit much. The quizzes are too easy and do not test much besides that you watched the videos.

This course is great for someone who has absolutely no knowledge of machine learning. You will leave this course feeling very confident in Linear and Logistic Regression. The only improvement that I would make to the course is to simplify the examples given in graded labs. I do not mean to make the graded labs easier, but rather to make the taught code more simplified. For example, in the last graded lab you are expected to loop through every w parameter manually and multiply against X[i] then add b after the fact. Once this is done, then you apply the result to the sigmoid function argument. The better way to teach this is to simply supply the NumPy dot product of vector W and X and add b directly into the sigmoid argument. I.E: sigmoid(np.dot(W, X[i]) + b)

While the content in this course is (mostly) rather basic, it is explained well with many examples that make it easy to absorb. It won't get you ready to do much actual ML implementation, but will set you up well to understand future material on ML. I have only two complaints with the course, both related to assessment. The quizzes and formative assessments are an absolute joke. They are all listed as requiring about 30 min, but I would be surprised if most people required more than 1 min to complete them. Conversely, some programming assignments include example "scaffolding" code in the graded cells that you are meant to complete yourself, but the scaffold bears no resemblance to the final implementation and serves only to confuse the issue.

The course has provided me with valuable insights and knowledge. However, I would like to suggest that future participants are made aware of the prerequisites. Familiarity with Python programming and a basic understanding of linear algebra are essential for a comprehensive grasp of the course material. This prerequisite information would enable learners to fully engage with and benefit from the content.

I appreciate the course's quality and believe that incorporating this prerequisite information into the course description would greatly enhance the learning journey for all participants. Thank you for considering this feedback.