I missed the coding parts

understood little bit

可能是偏向实际应用，对我理论方向的帮助很少

讲道理这么课内容有点少，而且有点空....

Very Theoretical !!

I very simple class

Its informational

就两周的课， 我不知道算是凑数吗

to theoritcal

too much bugs

No Homework!

Bit tedious

Bit boring.

A bit short

Too simple

Too short.

too easy

Too easy

okay

괜찮음

ich

k

Definitely not worth paying for (and I literally completed this in one afternoon). Thankfully I did not pay, so it was not that bad value in fairness.

In honesty the lack of value from this course actually says a lot about Andrew Ng's original Machine Learning course, which was consistently excellent. Actually coding in Octave for that class cemented a lot of concepts as well, which this course does not.

The title of the course suggests this is pitched towards more advanced students who already know about Machine Learning but maybe not so much about best practices. This feels far too basic for that demographic. The practices are sensible though and useful, if maybe overly focussed on massive datasets as opposed to the ones that Google *doesn't* deal with on a daily basis. Things like SMOTE could have been mentioned as well, for example.

TL;DR: This feels like a missed opportunity. My advice is don't take it if you've done Andrew Ng's ML course. Google things after that and wait for a decent course that's pitched towards intermediate students.

I found the course rather confusing than helpful. One of the key issues with video-only courses is lack of interaction of the user with the material. In previous Andrew's ML courses, this issue was cunningly tackled with "wake-up" multiple choice mini-quizzes. Such techniques would help the course a lot.

The questions in the exam were poorly phrased and full of typos; some had numerical issues (percentage of errors in the dev set did not sum up). Some of the answers seemed to contradict with the material as I remembered it from the course: f.e., the question on whether to get more foggy images to improve the model performance should have been answered with "augmentation is fine as long as it looks fine to the human eye". This contradicts to Andrew's remarks in the course video "Addressing data mismatch" video -> Artificial data synthesis. Are you sure we would not introduce a bias by adding artificial fog to frontal camera images?

Notwithstanding the great video lectures this course's assignments were poorly composed:

Firstly, there are no programming assignments! I understand the material here is mostly conceptual, however subjects such as 'Transfer learning' and 'Multi - task learning' should be given as a programming assignments. In 'Transfer learning' you need to modify an existing model, which I think is a good tool for a student. Hopefully we will use it in future lessons. Lastly some of the questions in both 'quizzes' have many complaints in the forum and the same complaints reappear yearly, therefor it's a bit annoying no measures are taken to modify the questions so they will be clearer.