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

I like the project perspective on ML tasks and the content a lot. I have two critics though:

1.) I am missing at least some smaller steps into the direction of implementing certain concepts (e.g. changing a model for transfer learning purposes)

2.) In addition, the videos are quite long, sometimes it seems as if the same audio/video sequence was added to one video multiple times.

Anyways, many thanks for this course. I think with some minor improvements it will reach the level of the other courses which are simply great. Many thanks!

Whereas the 2 courses that preceded this one in the specialization are focused on explaining how Deep Neural Networks work, this course is more for people with experience of NNs and how to troubleshoot issues that might occur in the wild.

I think the content here is really great, but if you're someone like me with no real world experience of Deep Learning, it is not so interesting as the other courses which explain the core concepts of the approach, rather than how to fine tune a real system to get better performance.

The curse is quite simple, there are a few interesting insights so it's not all bad. I feel I've learnt some interesting ideas. However, I feel it's quite incomplete. There are several problems that happen "in the wild" that are not covered. There is more that image classification and speech recognition to machine learning, therefore the experience of Andrew makes the course content biased to problems that are interesting but very specific. I was expecting something better given the quality of the first ML course.

While it was useful to see some of the best practices in ML, and the course contains practical information, the information could be delivered more concisely. Also, we get a lot of intuition, but the delivering of the material is getting less and less rigorous. The very least it would be nice to see some sources attached to each video. 3 stars may be a bit harsh, and it does not mean that I do not think it is important to listen to this course, it is more about the way of delivering the information.

As always Dr. Andrew Ng offers great insights into specifics of hot topics (Multi-Task & Transfer Learning) as well as providing unique "studies" as quizzes to complete each week. These quizzes are the primary take-away from the 2 weeks that offer a lot of redundant lecture material. Save some time... just make the 'simulations' the focus of the class then... perhaps use some transfer learning toward a different application in the quiz.

This is a decent course, but I found it less useful than other courses so far. There seemed like a lot of redundancy and repetitiveness in the descriptions, and I think all of the information could easily be fit into a single week that more concisely captures the exact same information. The quizzes in this course were interesting because it had a very applied nature (trying to capture real world scenarios you may encounter)

This is a valuable but misplaced course. After the first two courses, I expected to get hands-on experience with TF+Keras, and after that, or beside it, learn about strategies of tackling ML projects. However, by first talking about the strategies, one could miss many valuable points because one is not deeply aware of the necessity of these points. Hence, the course was boring comparing the last two.

Good class, but I think as part of the Deep Learning specialization that it'd be more useful if there were some programming exercises to reinforce what is taught in the videos.

Week 1 seems to reference a "flight simulation" programming assignment, but then it just has a description and a "mark as completed" button. Maybe this programming assignment is still being worked on or the content is wrong.

I'm grading this course lower than I graded the two previous ones for two reasons: 1) while there are many examples given in the course, it is actually hard to take a step back and see how to concretely achieve some goals in a more generic manner, and 2) in the assignments (which are made of quizzes), many "wrong" answers would actually be appropriate if more context was given.

This course really felt a lot more thrown together than the other ones, with a less cohesive lesson and quizzes that had more subjective material in them than usual. And perhaps it is a bit nitpicky, but I found the swipe Ng took at computational linguists to be kinda distasteful. I know there is a lot of bad blood between ML and AI people, but it has no place in coursework.

Thanks Andrew for this course! However, it is obvious that less care was taken for the preparation of this course compared to previous courses (more typos, etc). Some of the sentences in the quiz were not clear at all and made it very confusing to choose from the options. A little programming assignment at least would be nice to reinforce our learning of the materials.

The course goes through valuable practices when it comes to analyzing errors, and Andrew does a great job at explaining. Though, I felt that there should have been programming assignments to accompany the theory. I strongly prefer the layout of previous courses, "quiz -> programming" as that feels more interactive than just doing a quiz in the end.

nice lectures and very useful knowledge learned by Andrew, but it is really short and no working assignment through real code.... and quite a lot more mistake than course1 and 2. Really love the two previous courses, don't work why the quality of the course drop off so sharply.

Somewhat disappointed, but still really great lectures.

It gave much more industry driven approaches to improving the model. I as a student don't have that much experience with deeplearning and that' why I couldn't relate with most of the topics that were going on here. Of course, the teaching quality was supreme. But the course's contents itself felt a little bit dry to me.

The course work is really good. It has a practical emphasis. However, I did not like the quizzes (especially week 2 quiz) in the sense that the options are not very clear to understand and you end up being more confused. I hope the team works on the clarity of options for people who take it in future.

Even though the materials in the course are very interesting, I would expect that in the third course we would have more tools in order to work by ourselves in a project... It would have been amazing a final project where you can put together this tools. Nevertheless it is still an interesting course.

This course was less satisfying then the 2 previous in the specialization. A lot of repetitions, no programming exercices. Interesting test cases but feels a little out of scope because we have not done image and speech reccon yet. Consider putting the course at the end of the specialization maybe?

I think the week 1 was overstreched. There was not much content to deliver and for the first time Andrew's classes made me sleep. It was like the boring lectures we get at school. I think we can easily shorten the length of this course or just scrape it and add it to course 2.

I found this course to be a bit confusing with regards to what data set (training/dev/test) to fix under what conditions and so on. I've also missed having a practical home work, the case studies were fine, but I find that practical applications help me remember things better.

Some of the quiz questions (especially in the first week) were quite ambiguous. If I did not take the quiz directly after the videos, I don't believe I would be able to pass, Also some written summaries as in the 1st Ng's Machine Learning course would be helpful.

Some helpful advice here and there, but a lot of it seemed like common sense. It was not that difficult and a tad boring. Would maybe benefit from having us do actually data collection and cleaning tasks, or implement a ML pipeline and monitoring for the pipeline

Generally provides very good advice. Perhaps this course better placed at the end of the course as there isn't much hands-on experience involved and students would benefit form having experience with CNN's and RNN's prior to thinking on project-level scales.

Too many broad statements of "yeah, we generally do this thing for best results" with very little explanation of the background theory. I don't expect advanced math and derivations, but better intuition into why certain best practices exist would be nice.

This course was good, but it was pretty light on content to be considered a separate course by itself. Though the content is valuable, it could've been included as additional/bonus content on either of the first two courses in the DeepLearnign.ai series.

I didn't find this course as engaging as Course 1 -- there weren't any coding exercises and it felt like a bit of a let-down after the excitement of coding in Course 1. But it may turn out to have value when trying to start a new AI project from scratch.