学生对 deeplearning.ai 提供的 Natural Language Processing in TensorFlow 的评价和反馈

The course material is good.

However, multiple URLs are outdated both in the course material and in coding exercises (which makes some coding exercises not working).

Optimally some of the coding exercises should be updated to newer versions of TensorFlow (some things from the 2.alpha version are no longer available in version 2.4.x and some things are deprecated).

Also, it would be great if the coding exercises were graded (like for earlier courses in this specialization).

The content is good, the videos well paced. The code examples are also very useful.

But I feel the structure of the class is too loose. In my opinion, it would benefit from having assignments that must be submitted and graded.

Maybe they could be small and focused - like focusing on just working with the tokenizer, or setting up Embedding layers or LSTM layers. There could also be one where you load a pretrained model and writing the next token prediction loop.

I'm a bit disappointed with this specialization overall. I think I expected a deeper familiarity with tensorflow, more exposure to the TFData abstraction for large datasets, more low-level exposure to extending your models to fit a specific problem in your domain. Instead I feel like this specialiaztion would better be titled "Black box manipulation of the Keras API". That's a shame, given how solid the first deeplearning.ai specialization was.

This course is out of date and not as polished as the Deep Learning specialization. Data urls in the notebooks are broken. The quizzes are mostly random parameter names that you would google if you needed them, and the week 4 quiz actually has duplicate questions from week 3. The coding exercises are not graded. I did them anyway because I want to learn, but I also want to be challenged and want a certificate that conveys rigor to employers.

This could have been some more intense with 2 quiz in each week (1 or 2 tough questions), giving a written explanation of what a code snippet is meant for or each line of code is meant for, spend time on explaining fundamental concepts. Highlights of course, clear and crisp in explanation of concepts and functioning of code. overall, coherence is well appreciated.

The models developed in the course of the instruction were pretty useless. The instructor didn't discuss enough about how these models could be improved. The content of the course doesn't allow you to actually take on proper NLP and deep learning projects in industry. The demands of the industry are quite different from what's covered in this course

In my opinion, the course was too simple. There are many many concepts that are not covered properly. Even if they recommend going to the deep learning course from Andrew, I believe that at least could explain a bit more some parameters used in the functions and how actually work.

On the other side, you make cool thinks like text generation!

This course does not follow the same standards as the previous courses from deaplearning.ai. The material taught in this course are two basic and do not go in-depth to introduce the major techniques that are being used in the field. The colab notebooks are not provided in most cases and you have to look for them in QA or Github.

In the previous two courses of the specialization, coding exercises were compulsory and graded. In this course, all coding exercises were voluntarily and not well documented. It seemed to me that for whatever reason, the makers of course 3 (natural language processing in tf) put less effort into the making. Bit disappointed.

A good course overall, however the explanations offered on convolutions, LSTMs, GRUs were a bit poor. I know it is beyond the scope of the course, but it will help the student to know what an LSTM is overall and what is the meaning of different arguments (i.e., the `return_sequences` argument in LSTM class).

Basic concepts of NLP. I expect more from this course . Not helpful for real world problem. Should have add more content with more complex and real world problems with programing exercises. No assignments for evaluation of a student understanding. This is not expected from Deeplearning.ai.

It was a good course like the rest in the series, though in this course, they don't link to the colab notebooks that Lawrence works through in the items for each week. The colab notebooks exist on lawrence's colab account but you need to hunt them down. I would suggest fixing this oversight.

This third course provides main NLP concepts using Keras simple example codes. Just like Courses 1 & 2, there's no math and as explained in the videos, if you want a deeper understanding, then you want the "Deep Learning" specialization. Only quizzes, no graded exercises for this course

It was good but there are several errors in the code for some weekly exercises.

I wanted to raise a PR in the author's Github repo to fix theses. However, upon seeing the backlog unaddressed PRs in the author's Github repo, I didn't bother as they will probably not be looked at.

The length of the videos is fine. Personally, it bothered me that there were no programming tasks, the quizzes were too short and some of the questions were repetitive. I would have liked to see programming tasks, more quizzes and also intermediate questions in the videos.

Since the course doesn't mention "Introduction" to NLP, I thought that the course will provide a detail insights to Natural Language Processing but the course only covers basics of it. Also as far as tensorflow is concerned I was expecting more hands-on experience in it.

the lack of graded exercise makes this course somewhat messy. Many of the codes that are given to analyze don't work on the go. The back and forth between sets and cases of classification in my case at least, did not help to fully grasp what was going on.

week 2 and week3 are disorganized - the examples don't run without making modifications based on information in the forums.

However the overall course is worth it. I hope they pay more attention to making the examples accessible and making them work.

This course did not require lab submissions which I really liked because it coerced and helped me by providing an objective to learn more. It also introducted topics such as LTSM, Global average pooling and regulizers which I feel were too rush

Nothing against the instructor, he was really nice. But the content is extremely basic, to the extent that the whole course could be completed in one day. At least the previous courses had graded assignments, but this one was way too easy.

One example in case of text classification could have been theoretically worked out. For example classification using RNN/LSTM. How a word vector is passed as input to one unit of lstm? To view in on paper would make concepts more clear.

The course is good but lacks graded coding homeworks, these are the most powerful learning tools and it involves reflecting upon the matter, even if they have bugs or version mismatches, they are the most important learing tool!

Please give an explanation for each code in colab like in teh previous course. Moving between expalantion video tab and colab tab was troublesome. Moreover, there are some labs that contain an error which is predict_classes().

Good overview but the assignments seem a bit disconnected from the classes at times (e.g. when asking us to use regularizers). Transformers are a lot more popular now and they were not touched on in the course.

There was no continuity in the videos. Felt it could have been much better. And the level of course is too low. You should change the title to "Basics of Natural Language Processing in TensorFlow".