ワシントン大学（University of Washington） による Machine Learning Foundations: A Case Study Approach の受講者のレビューおよびフィードバック

This course is very frustrating because it uses a library called Turi Create that can't be installed on Windows 10. There is no support on how to setup you local environment after three days of frustration I decided to cancel my subscription.

When I opened the forum for week one all the threads were asking the same question: "How to install Turi Create on Windows 10."

It would have been much better if the course was done with a more popular library like Skit-learn.

This course is useless if you don't use a Linux or a Mac

The course includes some good, basic, information on machine learning. The instructors seem to know the material well. However, the exercises and coding are based on a python package written by one of the authors that, while free to students, does not easily translate into common packages such as Numpy, Scipy, Scikit-learn, Theano, TensorFlow, Keras, PyTorch, and Pandas. Also, the package used only works in Python 2 (which will no longer be supported as of January 2020).

I signed up for this course and began the reading and videos, but once it was time to begin interacting with the technology required (Amazon Web Services) , it appears this course is not longer supported by the instructors. Most communication on the course seems to have stopped between 1 and 2 years ago. Recent comments on the discussion forum no longer receive a response.

The Lecturers are very enthusiastic, but I was hoping for examples and assignments based on Pandas and Skikit-Learn. Instead the course examples and assignments are based on a machine learning package called Graphlab, that stopped working when it was upgraded to version 2 (there are workarounds that enable it to work locally, but clearly it isn't "enterprise ready")

I feel like it should have been mentioned a lot clearer before starting that this was simply a course in how to use the creators library. These skills are not transferable anywhere else as I would have to pay to use them in future! Would have been a lot more useful as a how to for sci-kit and pandas.

This specialization is delayed for months now - very annoying! Don't give them money!

i cannot spend $4000 per year package just to learn this course. sorry i am off...

Unsufficient information for the programming assignments.

Out of date. Should be retired or updated.

Not happy about course 5 & 6 got cancelled.

The course has a fundamental problem, it relies completely on a library developed by one of the instructors, which is not open source. In the end, it seems like a big opportunity of delivering a marketing campaign by the instructors then otherwise.

I definitely will not spend time and money on the other courses of this specialization.

The course could be great, if it won´t depend ob Python 2.7 and graphlabs (because scikit isn´t scalable). Also some quiz questions are so hard, that it is impossible to answer only with the material. So they use forum posts to answer how you can find a solution to the quizzez. So in total more a waste of time.

No Open Source Libraries, this course is not educational; it is a sales pitch to use their expensive software. Good luck having an employer pay this amount of money for software when they can hire employees that can use free open source libraries.

The fact that the class uses GraphLab instead of pandas/numpy/sklearn should have been stated up front

The course felt like an advertisement for the professor's toolkit

It was very disappointing that the equivalent standard workflow was not supported

The instructors need to specify that you can run this course specialization using MAC or Linux only. I have wasted my time for the past 3 weeks trying to figure out how to run the Sframe or Turi using windows and could not find any solution.

The instructions to download GraphLab don't work and even when you sign to use the AWS platform the instructions are also old and I haven't been able to start any of the assignments because of that! I want MY MONEY BACK!!!

Shame that it was not possible to progress with this course without using graphlab which the creator of this course himself created. Please see the course as just a training sales promotion for his ML application.

Very good overview of ML. The GraphLab api wasn't that bad, and also it was very wise of the instructors to allow the use of other ML packages. Overall i enjoyed it very much and also leaned very much

The course was well designed and delivered by all the trainers with the help of case study and great examples.

The forums and discussions were really useful and helpful while doing the assignments.

在上這門課之前，其實我就具備了這堂課大多數內容所需的知識，包含這些模型的方法以及數學證明等，因此這門課對我的幫助在於graphlab的使用、各種案例的探討及實踐。

由於有一些先備知識，這門課程的部分案例及題目，是我覺得不太能接受的，例如說：雖然課程中有提到overfitting觀念，但很多題目看起來都只在表達參數越多效果越好。

另外可能是在下才疏學淺搞錯了，在一些linear regression或是logistic regression的範例中，由於案例中的dummy variable過多，造成變數之間線性相依(n維空間中有k組向量，若k > n，必然存在若干向量彼此線性相依)，直覺上有無數組解都可以達到幾近0的SSE，因此縱使結果再漂亮，對那幾個case中的參數，個人其實感到相當的疑惑。類似的困惑還有推薦系統的上課實例等。

課程主要專注在案例分享及各種方法的簡介，整體順序安排相當不錯，兩位講師的描述也相當生動有趣，有很多地方讓人感到耳目一新、獲益良多。不過關於模型的限制覺得還需要更多的解釋，才不會讓人誤用了一些不恰當的方法。

I can only infer that this course's target audience is rich pregnant women who care about shoe shopping and celebrities. Unfortunately I am none of those things and had to cringe my way through the examples, watching the videos at 2x speed.

The course itself is incredibly shallow, even for a survey course, and basically serves as an ad for one of the professors' own products -- Graphlab Create. You'll be much better off taking Andrew Ng's course, which is significantly more in depth and forces you to write your own solutions to problems instead of relying on a proprietary library.

The only reason to prefer this course is if you really dislike the idea of using matlab.

I'm sorry, but this is just ridiculous. I can't recommend this course to anyone. It's all about advertizing: Emily Fox can't stop but recommend Amazon services, and Carlos Guestrin does the same for his Dato's Graphlab Create, which is might be great in general, but absolutely useless in educational purposes. The practice part of every week is just a waste of the time.

I can't say "money well spent".

LIKE an advertisment!

Why not use pandas and numpy sk-learn?

Open source is more popular!

not useful since the material covered lacks any rigor.

Might be good for someone looking for a casual overview?

I really wanted to like this course, and was excited about the series. Very disappointing. Refunded after scoring 100% on first three weeks and watching the theory portion of week 4. I was familiar, with the subject prior to taking this course; was hoping for a deep dive.

Too many trivially short and low information density videos. Handwavy mathematics. I would have liked to get a more solid idea of the depth of the series from the first course before committing money.

Default software for the course has near-zero market penetration (per indeed.com), unless maybe you work at Apple -- not really excusable for something that purports real-world value. Yes, you can use other software, *except for the capstone*, per another reviewer: this is fatal.

Presenters just not fully prepared to lecture on the topic: the nail in the coffin was the end of the week 4 lectures on clustering: "So, at this point, you really should be able to go out there and build a really cool retrieval system for doing news article retrieval. Or any other really, really, really cool retrieval that I can't think of right now. But of course there's lots of interesting examples. So go out there and think of ideas that I can't think of right now." Really? How about: "Cut! Take two."

Many poor design choices for the presentation. Too much time spent writing things on slides that should have already been on the slides.

As of this review: no reviews on the last courses in the series, and some poor (but indicative) reviews of the other courses.