I enjoyed this course too much, usually every company wants a recommended system, but the courses or examples available on the web are few. Very well explained many theoretical aspects.
Amongst all tensorflow courses this is probably the most useful. Using AI to make better and automated recommendations can benefit most businesses.
par Théodore M•
The courses is not up to date. They are using TF1 instead of TF2 and sometime python2. The WALS is not supported in TF2 and it is the central algorithm for collaborative filtering. I only learned that I should NOT build a recommendation system using TF,
One star, but not to content. But because the course don't have "Audit" option. It's mean that after subscription ended and you received certificate, You can't more access to video material in course. When subscription active, You can use mobile application and download video material for studying offline. Before yours subscription ened, copy video material to safe place for later review.
But the course content deserves a higher mark - 4-5 stars.
par Jesper O•
The labs by themselves - 'jupyter' notebooks - are good, but they were obviously developed in some other context and then reused in coursera. This is a problem. There about 6 labs per course - in each of the 10 courses of the two Machine Learning specialisations. Each lab starts the same way - connect to the google cloud, allocate a vm, check out a git repository - exact same repository for all labs. It takes 10 minutes. Not 10 minutes where you can go away and have a cup of coffee - 10 minutes where you have to be there and accept terms, answer 'Y' etc. If the labs are done outside the Coursera context you would be able to pick up where you left off in the previous lab - zero setup time. But not here - it is too much wasted time: 10*6*10=600 minutes. Evil.
par Liang-Chun C•
Not very intuitive explanation compared with previous four courses.
par Paulina M•
Overall a good and comprehensive introduction to recommendation systems.
On the downside, some functions used were deprecated, there was sometimes inconsistency between versions in labs (for example automatic upgrading to Tensorflow 2.0, which was incompatible with other libraries used in the lab and things like that).
Also, in my opinion an insight into the models' results is lacking. There was a nice explanation of the performance in content-based part, but later during hybrid and context-aware systems there was no comment on models' accuracy in comparison to the original basic solution.
par Quân V H•
you should move into tensorflow 2 instead of tf 1.x
par Ashley B•
The course lectures were decent, but the labs are full of bugs and erroneous or incomplete instructions. The final lab of the course (and also the specialization) has been unavailable for a few days now and tech support has not been helpful.
par Muhammad S U•
Most of the coding exercises use TF 1.X which is rather dated. I was hoping it would use TF 2.X instead. Also the coding exercises do not leave anything for the students to do. The video instructions were clear and well done.
par Nicolas S•
This one was the hardest of the specialization. A schema accompanying the code explanation would have been useful.
par José C L A•
The lectures excellent but the exercises were tedious and boring. Code only to be understand by the creator only.
par Sanjay K•
No tensorflow.. lot of talk not a single math.. NOt good
par Jakub B•
Very poor course. Assignments are very weak and they do not test anything - there is no grader, you can just verify solution by watching the lab videos.
The content is OK, but web is full of good content on recommendation systems.
If you want to take this course by any means do not pay for it - by paying you only get access to qwiklab platform which sucks for these kind of assignments, and anyway you can do almost everything from the course on GCP free tier, and also not lose your progress every time you log out of lab.
par Brice T•
Some labs with bigquery and the movieLens are not working - including the solutions, which is really time consuming and frustrating.
Labs illustrate very well the concepts and clarify the practical issues and solutions with gcp & tf. Excellent teaching !
par Dong Z•
Before the DAG part everything was quite understandable and useful. I am completely lost with the DAG and cloud composer part. Maybe it is not a very good idea to teach some tool that is still under developing.
par YUJIE M•
For this course, some part could be break into small chunk, and explain more detail. Generaly, this course is awesome!
par Mark Z•
Bad course overall. It has some theoretical content in it, but concepts are not explained in depth and videos are sometimes hard to follow. Speaking of assignments, I had no motivation for completing them, because, firstly, they are not graded, and secondly, they are terribly designed and one won't get much from them.
par David K•
Harder to follow than the other courses and did not love the teacher who led most of the lectures
par Kyle M•
content is outdated
par Manasi S A•
Don't enroll in this course because the qwiklabs and the videos are completely different the qwiklabs used bq while the videos taught tf , the solutions of the code lab isn't for the qwiklab , I think the qwiklabs are updated but the videos aren't , the final project qwiklab didn't the event use airflow and wasn't about a recommender system at all it was just a jupyter notebook , google please update the course content and even rectify the final project
The course is not updated. The labs are completely different from the titles and the lab solution videos. Quite disappointed.
par Rubens Z•
I love math, but unnecessary complexity was added to the content, making the course unpleasant to follow.
par Antony J•
The videos and labs are significantly out of synch, but despite this, I think this course (and the overall specialization) is a remarkable achievement. It gives you a high-level overview of content-based, collaborative, and hybrid recommendation systems applied to a real-world Google Analytics dataset.
Perhaps some of the lengthy code walkthrough videos are a bit misjudged, as TensorFlow is a rapidly developing ML framework, but they are worth persevering with because they illustrate what is involved in developing enterprise level machine learning systems.
par Sinan G•
Great work by Google, a lot of material and system walk-throughs. Apache Airflow / Google Composer is a smart tool but perhaps too complicated where more simple e.g. bash cron scripts could suffice - however it is understood that for truly scalable end-to-end systems the traditional single-cloud-virtual-machine solutions will not do. We are shown how that could look like and much more.
par Harold M•
This was a large and hard course on ML and in particular for Recommendation Systems. The videos were way to long. The content was very interesting. I've learned new algorithms like WALS for Collaborative Filtering and others more.
The Cloud Composer technology is cool for Keeping your System learning all the time.
Thank you Googlers.