워싱턴 대학교의 Machine Learning Foundations: A Case Study Approach 학습자 리뷰 및 피드백

good

nice

nice

cool

Good

Cool

nice

V

I did not have a lot of experience in machine learning, so this course was very good in the aspect of introducing people to machine learning concepts. Most of the times the material was very well explained, and I like the concept of the tutor writing on the screen at the same time they are presenting, personally it helps me more. Some of the quizzes were easy so you did not need a lot of preparation, some of them were more difficult or troublesome, like the quiz for Deep Learning. I also liked the graphlab module, I think that learning how to use it will help me with my own work.

However, as this course does not really go in depth in the algorithms themselves, I feel that after one month and a half I have a basic idea, but I haven't learned much about how to implement machine learning on my own even in basic things, while other courses have more or less the same time frame and are more dense in their material. In my opinion, this whole introductory course would just be just splitted and each of these intrductory weeks would be appended as the first week of the subsequent modules to come. Because anyways, after 4 months in the specialization, if somebody continues to the recommender systems module for example, he/she would have forgotten the basics of this so they would need to cover again the recommender systems week in this course. And from the other hand, if some introduction is again repeated in the subsequent modules, then why have this introductory course anyways?

Thanks.

The biggest downside of the course is that instead of learning on open source machine learning modules (sklearn) the course offers Dato's GraphLab, a proprietary piece of software that requires paid licenses to operate.

To be clear, during the duration of the course students can use a student license that provides graphlab for free but this expires after a year. It seems like fine software but if you arent going to purchase a license after the class expires whats the point? Also, Graphlab is built on top of python2.7. If you are running python 3.0+ on your machine youll have to install a python 2.7 instance.

Otherwise the quality is solid. The philosophical approach the professors take is to give you a taste of a variety of machine learning models. The upside is that if you want to get a taste you can. The downside the course feels pretty shallow and then the next course in the specialization -- regression -- feels like a pretty stark contrast. In general it could be argued that this is a problem with all coursera courses. How do you modulate course difficulty when you could be targeting students who are somewhere between high school kids to computer scientists? So the course and the specialization tilts between very easy and very hard.

The course is effectively a tutorial on how to use proprietary software to solve a range of machine learning problems.

I liked the fact that the course covered a wide range of problems quickly. There were however two issues that I did not like.

1) It is not well supported and given that the technology is proprietary, there are few other places that offer support (i.e. you can’t just look at problems and solutions on stackoverflow to get insight into the tech)

2) For a course labelled as “intermediate”, it presented very little detail. Most of the course was dedicated to explaining particular problems, the solution to which was inevitably “then you train this really clever, one-line algorithm we have written for you and you query it for insights”. I felt a little cheated by this approach to a subject which should be really fascinating.

While some of my concerns may be addressed in follow on courses, I am left with little insight into what really lies ahead. For example, is this really an “intermediate” course? What background do I really need? Will we ever get to the detail or will I always just be expected to call someone else’s brilliant algorithm and accept the result.

It should be 5 stars based on content - though I have a feeling it's a bit dragged out to create as many courses as possible for the Speclialisation.

However, I think IPython isn't a great tool for this, especially as the requirement is Python 2.7 for GraphLab as it doesn't support Python 3 yet. Going backwards, I think. I also think ideally you would want to encourage people to write bigger chunks of code rather than get bogged down with word counting.

However, the main issue is that the assignments are really badly put together. It's actually hard to understand what the underlying requirements are at first read - from about Week 3 onwards. The concepts are easy to understand, it's the way they are worded and jumbled. I had to read them over and over again because there is a fair amount of jumping around and since there is filtering of the data, the order you carry out the tasks matter very much.

I was looking forward to this course and planned to take 3 but now I regret paying for this one and thinking about finding another one. I think the tool and the messiness of the assignment make this Specialisation not up to the task, sorry.

I think this course is a relatively well put together, gentle introduction to machine learning. It would be good for people with zero experience with ML, who might be overwhelmed by other ML courses (Ng, Abu-Mostafa) out there. This course would not be useful for anyone who has any previous knowledge of ML.

Many reviewers have taken issue with the software used. I actually liked the Dato libraries a lot, but I'd be uncomfortable using proprietary software for research. I thought it was friendly enough to be appropriate for this kind of intro class, and I really enjoy iPython notebooks for interactive teaching.

I rated this course a 3 because of the price for the full course. This part of the specialization should be free. It does not provide enough instruction, practice, or content for the cost. Multiple choice quizzes are used to grade the programming assignments, so there is no feedback from the instructors. These answers are not made available to students in the free tier, which flies in the face of open learning. I am disappointed in the recent push to monetize courses. Please don't pay for this one.

Having almost completed this course, I can say that this course provides a basic over view of the concepts of ML. Quality and content are lacking. The professors, breeze thru the concepts in few short videos. No further guidance is available. You are on your own.

While the quizzes test the concepts learned, no live help is available. Again, you are on your own.

1. Some course materials should be provided other than basic fancy power point slides. The lectures don't do justice to the material content and quizzes. Some reference materials or write ups would be greatly helpful.

2. Teaching assistants or moderators should be made available for each course to help out with any conceptual or programmatic questions. Students are left to collaborate amongst themselves and figure out a solution. For a paid course, I expect some real professional help.

Many a times, I had to turn to google to help me out if I was stuck at a problem or didnt understand a concept. If googling was the only option, then I do not see a value in this course.

I doubt if I will continue the other courses in this specialization.

Thanks!

The instructors are great and the material logically detailed. The only problem is feed back or lack thereof. The assignments are hard for a person who has minimal computer skills as described in the outline, so having someone to go to for questions, especially the programming parts are essential and this is lacking unfortunately. Apparently, and I don't want to put words in someone else's mouth, but the assumption is that the fellow students will have the missing information and that they will actually answer them. That is not the case unfortunately. So even though I emailed the instructors with no response, I managed to obtain some information from other student who were extremely dedicated and much more advanced. I may have entered the course at a minimal personnel time which may account for my perceived difficulty. However, from some of the griping, I am not alone. On the other one gets what they pay for and I was overall glad to take the course and respect all the aspects except for the one already mentioned.

If you are already familiar with ML you won't learn anything new. The deep learning part is new, but it too short and lacks detail. I believe the concepts are explained in a clear manner but they are too high-level to be considered "learning". By the end of the course you will be familiar with some concepts of ML and the Graphlab API, but you won't be ready to implement anything by your own. However, I think this course is good to evaluate whether you like the teaching style and the overall style of the specialization. It would be nice to be able to skip this course and still get the specialization completed. if you already know what you want to learn and you don't want the the full specialization certificate jump into the other courses right away. I will continue with the specialization with the hope in the next courses the topics are covered in higher detail.

T​his course is a good high-level starting point. I particularly liked that the instructors cut the videos into very small chunks, making it easy for me to find good stopping points when juggling work and class.

My only complaint would be that it's starting to suffer from bit-rot: all of the videos are created with graphlab, and the notebooks with turicreate. The two libraries have very similar APIs, but they're not exactly the same, which leaves students occasionally having to google how to accomplish a thing with turicreate that the videos show working effortlessly in graphlab. Worse, occasionally the answers gained from mechanically running the class-provided notebooks with turicreate return different answers than what comes up in the videos using graphlab, leaving students to wonder whether they did something wrong or if the library isn't right.

The videos are engaging and the examples very interesting. But They use a library that only works with Python 2 graphlab) and needs some kind of environment not accepted by all laptops. I have 2 computers. On one I was able to install their library but my other noteboooks that are using python3 could not run anymore. It messed up my python environment and I can't get to clean every thing. I tried to install their library on another laptop (HP with slow processor) but the library didn't work. So I decided to use sci-kit instead. The issue is that don't get exactly the same results as they do. Which is an issue for the quizzes (answer for RMSE for example not the same) They should wait for graphlab to be compatible with python3 and to be less demanding in environment setting and to be compatible with normal laptop before offering this class.

The course is good. Very well taught. The issue is with Turicreate. Im not sure why we have to use Turicreate. Looks like its being promoted. Thats ok, but the real issue is installing it. I lost nearly a week on that. To figure out what to do and how to use it. It was so frusterating to start the course like this. But then, I found a solution in the discussion forum by Thet oo Zin, who has converted all the SFrames that can be used in Google Colab. That was helpful.

Then, many commands are in Turicreate. After this course I have to go figure out how those commands are written outside of Turicreate. For example:

1 - Load_Data = turicreate.SFrame('path of dataset')

2 - knn_model = turicreate.nearest_neighbors.create(x, y, z)

But the course is amazing, interesting and easily understandable for a newbee to python/ML like me.

I did pick up some very helpful information which was great, so for that I give it 3 stars. I failed to give it 5 stars because of the use of Graphlab Create and the subpar programming assignments. Apache Spark is a more powerful version of Graphlab Create, it's completely open source, and major companies like Netflix are using it. Carlos (instructor) is the owner of Graph Lab/Dato and uses this course to push and teach his platform. The programming assignments at times feel like he's just trying to teach me Graph Lab instead of the concepts. I'd have no problem with Graph Lab if it was completely open source, but it's not, so it feel like I spent a lot of money to be pitched Graph Lab. Class was not a complete waste, but I'd like it a whole lot better if they used Spark or open sourced Graph Lab.

I am giving this course 3 stars for a few reasons:

1) (Negative) Essentially no instructions were given for how to get Graphlab to actually work in Python outside of the notebook. I already have python on my computer, but the course basically only explains ipython notebook.

2) (Negative) I think the course would be a lot better if it didn't use this pretty graphical interface of ipython notebook. Why use this? I feel like this was done to dumb things down so that more people with no programming knowledge could get by. In reality it just makes everyone learn less. Using python normally, with graphlab imported, would be much better.

3) (Positive) The lectures on things other than ipython notebook were fairly good, and I like how the specialty is structured with case studies.

This course is a good intuitive-level introduction to machine learning. The presentation of the materials by the instructors is crystal clear and pretty much perfect. However, if you are looking (like I was) for a more in-depth course on machine learning, having already taken an applied-level machine learning course, skip this course and go straight for the next one in thsi specialization!

Big drawback also is the instructors reliance on GraphLab and related libraries. It is not commonly used and not really supported (for one, no Python 3 support!). I would strongly recommend making the required datasets for this course available in formats that accessible by libraries that are *far* more commonly used in ML applications like Pandas or Scikit-Learn

The course gives a nice overview of machine learning but does not go in depth. Of course this will be done in the following specialization but the pace might have been set higher to my taste. I also had a lot of trouble getting the software to run. As a matter of fact, the python package used (graphlab) uses outdated SFrame package which has changed name. FUrthermore, you cannot get the notebook running if you have installed anaconda3. A good thing about using graphlab is that it hides all the implementation away from the user so you can really play with the algorithms without getting to confused. A drawback is that this makes it harder to translate the knowledge to my own job where I do not have graphlab available.

Recently, I got a certification of Machine Learning course of Anderw Ng.

So the first course of Machine Learning Specialization is too easy for me.

But I think it's not a matter of how easy it is.

This program poorly explain how algorithms work

Even if the lecturers keep saying that we are going to study in detail in the later courses,

it's very difficult to stand boring situations.

And there's a serious problem.

They provide data for programming assignments, which shows different results compared to the one in the video lectures. So I am soooooooooo confused.

There are some small hardships more. But I am stopping writing this.

If at least one of the lecturers find my review, please contact me.

T​he good part is the very good and simple introduction of ML concepts, which are assesed during quizes. It makes you thinking and understanding the concepts of ML.

The less good part - t​he course should be updated in such way that video material fit the assignments. The usage of graphlab during presentations and turi create during assignments is not a good way to ensure complete understanding of materials.

T​he recording conditions were poor sometime ( police sirene or landing plane can be heard ). Some URL links on readings before quizes are not working and one assignment still has wrong answer considered as correct even the students posted clear messages toward mentors months ago.