學生對 华盛顿大学 提供的 Machine Learning: Classification 的評價和反饋

more topics like deep learning, neural networks need to be introduced

Good, but overemphasizes niche software product (graphlab).

The exercises/assignments are far too simple

too simple

good

.

I leave 2 stars as I learned a lot of new information and methods, and the theory and math behind them.

You will learn about Data Science and Machine Learning, but not much about Python.

The course is pretty much abandoned and outdated. Sframes and Turicreate packages (instructor's creations) are used instead of more universal packages. Installation in the beginning took some time and research. Many of the assignments have errors and bugs in the code that have not been updated. Forum assistance is abysmal for clarification or deeper questions. Many links are dead.

There are many times in the lectures where the instructors are writing several sentences in their handwriting on their notes instead of having the text ready to appear.

I would suggest using this course and series as a supplement to other information one as learned, not as an introduction for initial understanding. I found myself frustrated too many times.

The video content is awesome. Important concepts are being clarified in a very simple manner. However the evaluation method really sucks. First, there is too much spoon feeding in the programming assignments, which was not the case in earlier courses in the same specialisation. Secondly, in a few assignments, the answer to the quiz questions are sensitive to the platform we are using (like PC vs AWS instance). This was really frustrating given that the issue is known for a long time and has not been fixed yet. At the very least, there should be a warning on the quiz page itself.

The assignments are well thought out and explain the algorithms step-by-step. The subtitles/transcripts are a disappointment :( . Full of mistakes. Sometimes to the point of being useless or even worse - saying the exact of opposite of what the lecturer says. Since the lecturer sometimes is unclear - this is problematic. As usual - Graphlab Create sometimes crashes, however there are explanations how to run the assignments using Scikit-Learn.

The content seems rather thinner than that of earlier courses in the specialization, and seems to get more so as the course progresses. (Week 6 is entirely spent on Precision and Recall, with only about 30 min of lecture.) It feels like there was a rush to get the course out and that corners may have been cut at the end.

And as others have mentioned, several very important classification topics are conspicuously missing.

Overall good explanations in the videos; however, too much reliance on GraphLab, so that it seems more like promotional course for the instructor's own software and company. Also, the course is generally a bit light on content - the only algorithms discussed are Logistic Regression, Decision Trees and AdaBoost. Spending a full week on precision & recall is way too much time.

This course needs to be re-created using new professors.

Way too lazy IMO.

Too many "trick questions", total confusion between Python 2.x and 3.x

Too theoretical, almost no practical examples

Quizes are very poor and give no "hints" or true workhtrough examples pror to test.

This is a problem with all Coursera, though.

I am afraid that this course is, to a large extend, a marketing tool for promoting the instructors' proprietary product. Its use is therefore limited for the practitioners that want a foundation on the free Python data/ML capabilities.

I would not recommend this course to my colleagues.

Not as polished/comprehensive as the previous courses (especially week1, week5 and week6). But useful techniques nevertheless.

The speaker is very difficult to understand, and the environment for writing code is awful (web browser).

Extremely highlevel, quality of the material is significantly lower than in the previous courses.

Course is too slow to keep focus, it's repetitive but not clear when it's really needed.

The classes are not practical, and the voice of the teacher is very monotone, boring.

Explaination Is Not good I have to take help from other courses

worthless

I enrolled in this specialization to learn machine learning using GraphLab Create. Half way into the specialization the creators sold Turi, GrapLab's parent company, making it non available to the general public (not even by paying) and then all the knowledge devalued. I wish I had known this and I would have enrolled on a different specialization. The creators still give you the possibility of using numpy, scikit learn and pandas but I had already done a lot with GraphLab create. The time I invested on my nights after work became a waste. I was trying to convince the company I worked for to buy licenses for GraphLab create.

Coursera should not allow folks to create courses that promote a private license course because it would make people waste their time and money if they decide to privatize the software.

Don't take this course, and if you take it then only use GraphLab create when the authors give you no other option.

Teaching style: Carlos was good, Emily is not very clear and loses focus of the topics and often rambles. She seems very knowledgeable but she lacks clarity of exposition when compared to Carlos or Andrew Ng.

I was pretty disappointed with this course. Firstly, the course did not seem well balanced meaning that some weeks--particularly week 2--had A LOT of materials to watch and really felt like it was two weeks crammed into one, and then other weeks barely had anything.

Secondly, the exercises seemed unclear, poorly thought out and not really helpful. There were many errata that really should have been fixed in the beta iterations of this course.

Thirdly, I really would like to see more application and less discussion of implementing algorithms.

Fourthly, the "scaling" section was also a major disappointment. While it is mildly interesting to learn about stochastic gradient descent, I think it would have been more interesting to have a discussion about how classifiers work in a parallelized computing environment or actually to try one out using Spark.

Finally, given that GraphLab/Dato/Turi was just acquired by Apple, I question whether it is worthwhile to take this course as ALL the materials are taught using a library that in all likelihood will cease to exist.

The videos are fine. But, It's SIMPLY TERRIBLE to force people to pay to be able to do the quizzes. There was no such a thing in the first two courses (by the way, I gave high rates for both). It is OK to pay for the verified certificate, however, disabling the functions in the course is a wrong way to earn money, because people who want to learn the course might not necessarily want the certificate, and this is unfair to them because it limits the resources available. This whole Specialization thing starts to make me feel like you guys are in urgent need of money, rather benefiting the community. Remember there are tons of free resources on the internet, and this only undermines your strengths. You will lose tons of potential fans. Stop being seemingly arrogant.

If you are considering this specialization I would recommend the Andrew Ng course instead and the main reason is that it isn't depend on proprietary ML framework. Despite the good lectures, the assignments don't help you develop the knowledge required for ML developer role.

Taking in consideration the permanent postponing the courses delivery, from summer 2016 to summer 2017, finally the most interesting part of the specialization was cancelled. I'm completely disappointed with the specialization learning expirience.

very Vague and in efficient in transferring the knowledge. Teachers have tendency to overcomplicate very simple ideas to look more mathematically in-depth. It is not true and just causes confusion. I ended up to look only on slide and do the exercises rather than watching their videos