Graded exercise would be nice

No code challenges - 4 stars

It is wonderful experience

it's good for beginners.

Very useful course

Awesome experience

No graded exercise

C'était géniale!

it was good

Good course

Okay

I am a software engineer with a good base knowledge of machine learning and neural networks, and I took this course to get more specific knowledge about time series and TensorFlow to help with a project using stock market data. The content of this course is very shallow. I don't feel like I learned much reusable knowledge because much of the course is basically walking through code in Jupyter notebooks. If I wanted to just learn to copy someone else's code, I can do that on my own (for free) reading blog posts or tutorials. Also, quiz questions that ask about function names or names of libraries do not show any understanding of concepts and really just felt like filler because they needed 10 questions but hadn't taught any concepts to ask actual questions about.

I'm giving this 3 stars instead of 1 because maybe the audience is supposed to be students with less knowledge of machine learning or programming, or maybe it just doesn't match my learning style.

The course in general is good and introduces you to the uses of tensorflow keras API with different cases, but i can't give 5 stars because i think it still lacks on fundamental teaching about tensorflow.

I mean that during the course some tensorflow tools appear out of nothing, mainwhile i think would make a lot of sense to dedicate at least one course's module to introduce tensorflow library itself.

Just an example: during the last week we make an extensive use of tensorflow "Dataset" class to load the data into neural networks, and this tool appears out of nothing, but it seems very important and useful stuff that i think would deserve some introduction and explaining before you use it.

This course is great for introduction. BUT it is still lacking very important background information of the Tensorflow Dataset and how to master it.

It makes sense not to go into too deep on the NN model and their theories but when it comes to practical usage of any machine learning packages, data pipelines play very significant role (count towards 60% - 70% of the codes).

In the course we briefly talk about Dataset and use only a few APIs without explaining why and the logic behind them. And tutorials from tensorflow's officials still lacking useful guidelines when dealing with dataset of multiple dimensions.

I found the start of the specialism was very well explained; and as a result now I really understand CNNs (as it is was explained much better than the other courses I’m doing on Udemy and LinkedIn Learning). However I would suggest that Andrew and Laurence revisit the latter part of the course from a learner point of view, looking at the pain points along their journey through Sequences and Predictions. Overall, the structure of the whole specialism can be improved, and I find it not as good as my previous course (Andrew’s Standford University Machine Learning Course which was brilliant)

Though the course is very educational, the prediction is done at the right way. One can not use the windows of validation data to test it. The testing accuracy should be measured via point by point prediction which predicts the future value based on the predictions. At that way, the hardness of the problem makes sense, otherwise anyone could use the linear regression models rather than LSTMs. Please review the content again since it requires lots of stuff that is not covered like multivariate analysis, sequence prediction as well as point b ypoint prediction.

This is a good introduction to the API of keras, but that's not what I would expect from a "Tensorflow In Practice" Specialization. This is really an "Introduction to Keras" specialization, and really theory light one as well. As a graduate of the Deep Learning specialization, I expect this to be a way to apply that theory to large datasets and to novel architectures requiring some leverage of the lower level tensorflow APIs. Although I thought this course was well made, I feel it was not ambitious enough for it's name.

I would leave 5 stars for the instructor. But the support you get from the forum sin minimal. There are tons of recurrent, important posts and threads unanswered...some of them even have months old. I may have posted or upvoted ten different questions and maybe received answers for three...and from fellow students who may or not may be right. That could really seem like a good place to start looking at some improvements. Not to mention the constant workarounds you have to do to successfully submit assignments.

I echo some of the comments of others. The code needs to be more commented with explanations. There were details in the code that were not mentioned in the lectures or explained. When someone does not understand a particular line, then it is difficult to understand the rest of the code. The Deep Learning Specialization was much more complicated than this specialization, but I understood it better because it covered more of the details with clarity. Much of the code in this course had no comments at all.

Difficult to rate this course. I'm definitely glad I took it as it gives a valuable information, but it could've been much better. It should not be a 4-week course as it takes 4 to 8 hours of time to complete. Most of real world time series are multivariate, so I was expecting the course to touch them at least a little. Also I'm not a fan of the "if you do everything exactly as you've seen in the videos, you will pass" style of assignments - I'd expect at least some improvisation to be required.

Too much focus on creating synthetic data and arbitrary code. Unlike the first three courses this was hard to follow with significant gaps in the material not explained.

Although I passed I am still unsure of what I have learnt on this course.

My advice would have been to use a real dataset from the start and build on this and eliminate all the helper functions that just really proved a distraction. This would also be a great motivator if the dataset was interesting.

Just like Course 3, Course 4 was a let down. The content is interesting but I think unlike Courses 1 and 2 it is presented in a way that is kind of plain and not really all that engaging. I also think the assignments should still be required as this adds incentive to really work hard at it. I learned a decent amount, but Courses 3 and 4 of this specialization were a disappointment.

The materials are outdated and they are not as described as the first 2 courses (the effort and quality to make the curse is not the same as the others). The notebook from the first courses are very well documented and the ones from the last two are just code. Other than that the curse is great and well explained

Much preferred the Andrew Ng courses where we spent time coding specific sections of various neural nets. This ourse was practical, I guess, focusing on the mechanics of prepping data, but I don't feel it helped my understanding of the various machine learning techniques at all.

Some of the topics could have been described in details. There was always some kind of rush going on. By the way, I have come across several datasets and those labs introduced me to various neural network and their application using Tensorflow and Keras. Thanks to Laurence.