Comentários e feedback de alunos de Structuring Machine Learning Projects da instituição deeplearning.ai

Course contents are great as it talks about how to improve performance by giving real world example. This is one of the most crucial pieces in any model building task, but still is less focused in traditional courses. Andrew Ng's team has dedicated a full course on this aspect, which I believe will do the learners a huge benefit!

The quizzes were a little annoying to get through, as it is not much about deduction or reasoning, instead it's about learning the advice or rules mentioned in the videos. I think an actual implementation of a learning project and applying the error analysis, transfer learning, etc, would be more beneficial for the student.

Gave me a broader and more strategic perspective on how to structure and run a Machine Learning project.

I just felt this course came too early in the learning process. It would have far more relevant and useful had it been a more downstream course.

This does not take away from the fact that the content is very relevant

This was a good course because it "kind of" prepares us to real world projects and we think about what to do when different problems arise. I would also really like if this course included a section on how to create datasets from images, sounds etc. and prepares us for the "boring" parts of machine learning as well.

Was extremely helping in providing ideas on how to start and work on machine learning projects. Provided clear and well thought out ideas on how to make the most use of time and data. A small improvement can be made to the course by dividing some of the contents of the course to another week for better structuring.

Course is great. All concepts are explained very meticulously. Lots of respect for Andrew NG. Just a small suggest please don't give more examples on cat classification. Autonomous driving case study was good, speech recognition examples are good. Please give more realistic examples, that can be used in interviews.

Great explanation on how to structure your machine learning projects like distributing data among train & dev/test set then what to do for each type of errors to continues to transfer learning, Multi task learning, End-to-End Deep learning. It has been a fantastic journey learning about these different techniques.

There were definitely useful pieces of information in here, but I think it could have been condensed and delivered as part of the previous course. I liked the flight simulator quiz approach. Sometimes the wording of the questions was tricky and that may be causing people to get stuck even if they know the material.

-You should edit better some videos, in some parts Andrew repeated what he said, or there were long silences, or also what he was writing wasn't in tune with what he was saying.

-I'm not sure if the topics covered here justify a whole course. Maybe the insights shared here could have been inside some other lecture.

The flight simulators' results were not consistent with the advice provided in the lectures. I'd suggest being either less black and white in the simulators' answer responses, or, being more polarised (more black and white) in the advice provided in the lectures. Otherwise, this is a 5 star course. Many thanks!

I liked the course but it contained quite a few glitches which could be easily removed to improve the overall experience. E.g., once Prof. Ng makes a long pause and says "test". Sometimes the same ending is placed twice or in the final "Heros of Deeplearning" video Prof. Ng seems to ask the same question twice.

this course is pretty different from other courses in this specialization. It gives high-level knowledge of machine learning instead of implementation details. The course content is useful but it seems a little boring to me because I can't do any fancy, real machine learning projects as exercises in this course

The quizzes were actually a great idea. The content is definitely useful, as I've had similar difficulties in my company. I felt the videos took much more time than they should, though. A lot of the content could have been resumed in shorter videos. It was the first time I actually had to accelerate the speed.

Good materials.

This course was really short though. It seems to be a bit artificial to make a "specialization" out of these courses, where they could easily also fit into 1 longer course. Fortunately the dates you can start the courses are flexible enough that you don't need to wait (too long) between courses.

Excellent information about how to diagnose errors during machine learning and complete projects well. I would have liked a small coding aspect to see how certain concepts (eg. train-dev set, transfer learning etc. are implemented), even some very basic examples would have helped. Overall still a great course

Good course to get started with Machine Learning, the introduction video could have used simpler languages though as many of the jargon might not be familiar to newbies (therefore scare us off!!) and they are really not necessary prerequisites to the course. I enjoyed the quizzes as they are real and useful.

Very interesting material covered - not too many courses have this kind of information.

A little too short and very no practical assignments (only quizes). It would be very useful (although I agree quite time consuming to prepare) to have some programming assignments that deal with the topics in the curse.

Good general introduction to analyzing errors and avoiding common mistakes in machine learning projects and some info on transfer learning and multitask learning. Could've used references for further reading. It should emphasize exploratory data analysis and an ethics review as the start of any project.

4 stars for a very good course that should be improved. Course is still good, but it is not as polished as the first courses in this series. I rated those with 4 stars, too. There are mistakes in the quiz names, grammatical errors in quiz questions, etc. Never-the-less, it is the best of its kind.

Good course helped a lot to gain insights into the problems of machine learning but I would say more exercises are needed even if these guided exercises are good.

maybe add an exercise where you can simulate a fake NN and get the result and then choose what to do to get a feel of what you should do.

Interesting insights.

The insights could be visually structured a bit better so that I can also check them after the course as a reminder.

Often recommendations like if then could be put in processes or cheat sheets

overall: very valuable course regarding the insights and encouraging style of Andrew Ng

Well structured course, Andrew always never lets down your expectation, the explanations were very clean with the best appropriate examples to suit the explanation. Being more of theoretical, the task of giving us the correct intuition is really well handled by the way the lectures are structured.

Instead of clearly and concisely formulating some of the ideas (like F1 formula for the metric), they are discussed as if they were hard to understand. This makes the understanding harder and is rather annoying. The assignments are useful - they make sure you understood the material the material.

This was a good intuition course. I learned a lot and loved the content. However, I am afraid the information here needs to be repeated many times to make it a habit (as part of programming exercises). That's why I am giving it a 4-star. I feel like this could have been part of the last course.

Great overall. However, a major thing that is missing is the different between val set and dev set, and about the recent trend to perform K-Fold CV on the training set to get the val set. Maybe still need a separate dev set because of the different distribution if it comes a separate soure?