- 4 reviews
- 4 completed
I missed the first two sessions of the course because I didn't know what Machine Learning was and figures it was some theoretical topic with little practical application. Then over the last few months I have been reading and hearing a lot about Machine Learning and how it is being used everywhere and hence I decided to take it. This course is fairly mathematical but the math is easily accessible. But Prof Ng's lectures are amazingly clear, precise and easy to follow. The homeworks allowed us to solve real practical problems that the machine learning community at large solves. I used the Neural Network we learned in the class for digit recognition to enter the kaggle competition. This class made a lot of cutting edge (and not necessarily easy to understand concepts) fairly easily accessible. It is easy to see that Prof. Ng really knows this stuff. It is obvious that there has been a LOT of work that has gone in preparation for this class and I am very happy to have taken it.
This is a great course. I have the following suggestions for your course: 1) Slides have too much text. Please add more graphics, animation, movement etc on the slides. Even if you actually take a pen and write stuff (underlining, drawing rough figures) helps break the monotonicity of the slides. It helps me concentrate and draws my attention to the part of the slide or concept you are talking about. Online courses can be made more "interactive" if it isn't just a slide accompanied with a voice. 2) I think the course organization can be changed to better tie up the concepts more tightly. Most of the time I was going through weekly lectures as stand alone units and it was only when attempting the hw questions that I had to think and tie the concepts together. 3) Finally this course needs SQL background. You should add a lecture giving some background about SQL in the first week. 4) Any reason why you used octo.py and mincemeat.py ? There seems to be a mapreduce.py package that is very intuitive and easy to use. Finally please add some more programming assignments that work on real life projects, data analytics, web analytics, big data. The course currently seems too theoretical. Thanks for your hard work. Please take these suggestions as positive feedback. Appreciate you taking time to teach this course.
Its true. The professor does talk very fast. But this is the first MOOC. I think the professor will slow down next time. I took this class to learn more about statistical/data analysis concepts. The professor does a touch on a lot of stuff but does not go into depth into most of them. However he is very very accessible and approachable. He participated in almost all the forum posts and was very helpful to students. One of his research associates also took the course with us and helped us with homework questions when we were stuck. The forum quality was exceptional. The course cohort was very small (only about 700 people or so were completing the course at the end) but most of them were doing very well in the quizzes (and completed all 8 of them even if they needed to complete only 6 to get 100% points.) This should tell you how involved they were. The professor also has a great sense of humor and is obviously very well known and well regarded in this particular field. Anybody curious about this field will do well to take this course. And if you decide you want to learn in depth you can do a master's at his school.
This is a great course. I did not have any background in the material and it was fund to learn a lot of the concepts. If you know basic linear algebra (and little bit of vector math) this class shouldn't be too difficult. Joe Konstan has been in this field for a long time and is well known in the field (a lot of papers reference joe's papers). The workload is on a larger side, which is true for most of the computer science courses as well. A distinction about this MOOC was we were taking it in parallel with university of Minnesota students taking it for credit at their school. As a result the course was setup to be a normal "university level course" together with timed midterms and final exams and a lot of written and programming assignments. Studying for the exams and all the assignments took up a lot of my time (the course itself is not hard to follow/understand) I have the following suggestions for improvements: 1) De-couple lenksit from the assignments. I did all my assignments in R. The DB is fairly small that it can be easily done in R. Also structures dealing with matrices and dot products and vector magnitudes should be done in R (which is optimized for it) than java. If you cannot/do not want to decouple lenskit please add one more week of instructions to go over java concepts and lenskit concepts (which you already do) that will be needed in the course. There is no need to know java. People knowing any programming language can take this course. 2) The course is all encompassing overview of the the field, combining technical aspects together with business considerations. I appreciated that very much. Whats the point of a recommender system that is technically sound but doesn't do a good job of helping a business meet its need ( i.e RMSE is low but lift is low as well). I feel however there was too much static text on the slides. the slides can be made better to have more figures, more movement etc. I appreciate the fact that in some of the videos the professors derived the formula on screen with hand. That made it less mundane. 3) Finally it seemed that the class needs to be freshened up. While we did have a lot of interviews with experts in the field (I particularly like the interview with Netflix and LinkedIn since that is a contemporary use of the recommender system by companies to actually sell their products and make money) it seemed that we were reading papers and learning algorithms that were 10 years old now. Should we be learning something that is more contemporary? Thanks a lot nevertheless. I enjoyed the class a lot and appreciate your efforts (obviously a lot of it) in putting the class together. I learned a lot as well an would recommend it to anyone interested in learning about recommender systems.