I just finished the Machine learning course from Coursera, taught by Andrew Ng. I really enjoyed it, and I think I learned a lot. I would happily take another course from Ng.
However, my impression of Coursera is that they basically don't give a shit about quality. The videos don't play on Chrome on Ubuntu. There were occasional uncorrected disfluencies -- cases where Ng started a sentence, stopped, then restarted the sentence; clearly, these should have been cut. And the last assignment had two bugs. Neither were showstoppers, but both should have been corrected. And yes, people did point these out on the forums, with no reply from Coursera. If these were the only issues, I would chalk them up to minor growing pains.
However, the Computer Vision course was a clusterfuck. I found myself sitting there while the professor silently wrote things on the whiteboard. This makes no sense -- *software* could have cut that shit out. Or like $50 on mturk, at worst. Also, the pace of the course was too slow by far. In a ten-week computer vision course, you cannot spend the first week proving minor results in geometry (with lines like, "if you remember this from Freshman Physics"). This is really too bad, because I have a really neat computer vision puzzle concept which I would love to know enough to be able to code.
Finally, having completed the ML course (that is, aced all of the review questions and programming exercises that remain open), I think I ought to be able to click a "Get my motherfucking certificate now" button, even though the last video doesn't officially air until next week (they were releasing them a week early), and the last assignment isn't due until like August. I'm done -- let me out of here!
This entry was originally posted at http://novalis.dreamwidth.org/171672.ht