A locally organized online BDA course on G+ hangout?

Posted on by

Eoin Lawless wrote me:

I’ve been reading your blog (and John Kruschke‘s) for several months now, as a result of starting to learn Bayesian methods from Doing Bayesian Data Analysis [I love the title of that book! — ed.]. More recently I completed a Coursera course on Data Science.

I found learning through the medium of a online course to be an amazing experience. It does not replace books, but learning new material at the same time as other people and discussing it in the forums is very motivational. Additionally it is much easier to work through exercises and projects when there is a deadline and some element of competition than to plow through the end of chapter exercises in a book. This is especially true, I believe, when the learning is for a long term goal, rather than to be used immediately in work, for example.

My question: you are obviously evangelical about the benefits that Bayesian statistics brings, have you ever considered producing a Coursera (or similar) course?

And Flavio Nardi writes:

I am a PhD student in finance with a strong interest in Bayesian statistics. I have never taken a course in Bayesian stat before, but I did buy the Bayesian Data Analysis book for self study. Though very well written, I feel that supporting lectures would enhance the learning experience a great deal, of course. I have been looking for a serious online course dedicated to Bayesian statistics, but found none. I wonder whether you have considered teaching a free online Bayesian statistics course based on your book, maybe on Coursera?

My reply:

I’m not quite sure what it takes to put together such an online course. If someone came to me with a plan, maybe I’d consider it.

In the meantime, I’m trying to set it up so that my course this semester at ENSAE will be a G+ hangout (each Friday 8h30, starting 18 oct). I will post the slides and homework assignments. If you’re not in the course, we won’t grade your homeworks.

But maybe those of you who are interested in taking the course remotely via G+ hangout will be able to organize something where you grade each others’ papers, or hire a local “teaching assistant” to do the grading for you? (Then these people are the ones who make the money from the course, but that’s fine with me. Maybe that’s a more sustainable model for such courses in any case, rather than having many thousands of people all pay to a central organization which then must hire TA’s etc.)

25 thoughts on “A locally organized online BDA course on G+ hangout?”

  2. I’ve also been surprised by the lack of a Bayesian MOOC. Maybe it’s just hard to automatically grade the work, and there isn’t trust in the ability of students to accurately do peer grading.

  3. I am an economist (non-academic) and read your blog often. If you did an online course, I would definitely sign up. Four suggestions (if you do it) :

    i)Don’t rush through it. Do each lecture as you would in a real classroom.

    ii)Be relaxed about grading. Some people take these classes for the certificate. But I have a doctorate and several others like me just want to learn. One meaty assignment a week is more fun and more useful than a lot of little problems. Perhaps even something that has an unsolved/unresolved component to it (as the last bit of the exercise). This last bit will have to have a prediction (and/or objective) element to it, in order to automate for scale.

    iii) Let it coincide with (or follow) the release of the third edition.

    iv) Two 13 week stretches.

    What’s in it for you : a) influence b) book sales c) feeling good about sharing your insights with others

    I think that you would be a fantastic MOOC teacher.

    • Econ:

      Right now I’m not planning to create a new course. Rather, I am planning to put my existing course on a G+ hangout. But I was hoping some of the grading etc. could be organized by the remote students, as suggested in my post above.

  4. Will the slides, homework, and hangout be in French or in English? Just curious.

    And to Dave: maybe there’s a simpler explanation. Doesn’t a Bayesian need a prior (MOOC, that is) to get going? :-)

  5. I found it very hard to understand Bayesian Data Analysis (first and second editions) as an outsider w/o a strong background in math stats. One of the problems I had with BDA were the assumptions it made about things I already knew. For instance, the first hierarchical modeling example in chapter 5 assumes you know moment matching and can reason through a change-of-variables for the prior lickety-split. Those two moves mystified me.

    So if there’s not already a good lower-level intro course, I think that would be the place to start.

    I like computational courses, myself, being a computer scientist by training. Grounding out discussions in working code neatly ensures no details are left out. BDA leaves a lot to the reader to fill in, which is great for the level it’s at, but not so good for an intro.

    What actually got me over the hump in understanding was Gelman and Hill’s Regression … book. It grounds out in code I could actually run and understand. It goes into some algebraic details from time to time, but doesn’t assume much (any?) calc or much linear algebra.

    As a more theoretical intro, I like Peter Hoff’s book, though I have to disqualify myself somewhat because I’ve only skimmed through it and when I did, I already knew all the material pretty well. (I’ve been trying to help others learn Bayesian stats.)

    The Stan development team has talked at various times about putting together a short MOOC to introduce Bayesian stats using RStan for computation.

    But it looks like a lot of work to do it well. If it were just organizing the material and doing lectures, that’d be relatively easy. So when Andrew asked if somebody could come to him with a plan, I’m guessing what he meant is that if someone else wanted to do the organization side of it (fiddle with platforms and video cameras, organize assignment grading [crowdsourced or expert or a mix of both], and so on), he’d be more inclined to pull together course materials and lectures.

    • Great comment Bob!
      It is really encouraging for some of us who struggle through these things that the author of STAN had the same struggles.

    • As someone from computer science I have to agree that the schools example in Chapter 5 (assuming you’re referring to it) is confusing. I’ve had a lot of trouble trying to understand exactly what was going on there.

  6. A compromise solution is just to “slidecast” your lectures. My workflow for making slidecasts is very easy, but it relies upon Keynote. But it does work fine with beamer PDF talks.

    These are the steps:

    (1) Import a beamer PDF presentation (use http://www.cs.hmc.edu/~oneill/freesoftware/pdftokeynote.html)
    (2) Record the audio and slide advance timings during the presentation (start your talk with “Record”, not “Play”)
    (3) Later export a movie of the slides, synced with the audio (File/Export/Quicktime)

    All this requires is that your presentation computer have OS X and Keynote. Make the presentation with beamer still, on any platform you prefer.

  7. I’m looking forward to the October lectures, and the delivery of my copy of BDA3. Having said that, I think that for several reasons the model that Coursera has developed is hard to beat.

    Within a few days of starting a course people know what to expect in terms of how lectures and assignments are handled. There are forums for students to ask each other questions. It also provides a central place to browse for online courses. I highly recommend enrolling for a course and trying it out for a week or two. (Andrew Ng’s Machine Learning course is starting in October, it might be of interest and appears to be highly rated).

    Exercise and assignments are a key part of the experience and as I noted above, doing them ‘live’ with other people is definitely different to working through examples from a book. It does raise the issue of grading though:
    1) Programing based problems easy to grade automatically
    2) Essay based exercises seem to be quite amenable to peer grading. The Data Science course I took had a exercise around visualization, and peer grading was actually a good learning experience in itself. I imagine this is true in general.
    3) Exercises that involve developing tricky abstract arguments or that deal with subtleties may not be easily graded either by computer or by peers (unless perhaps the grade recieved was a weighted average, with the weights based on the previous scores of the graders involved).

    A friend of mine works at a UK university. I was raving about Coursera to her and she subsequently tried one of their courses. Later again her university investigated the costs and benefits of developing online courses . It appears creating a course can cost in the region of £14000, accounting for equipment and staff costs. The university, not Coursera pays for that. The teacher provides the expertise and time, and receives some benefits such as the chance of reaching thousands of students, an opportunity to promote a book and increased visibility. Is that a fair exchange?

    Finally, I’m not sure there is a feasible halfway house between simply making slides and video available and going down the full Coursera/EdX route.

  8. We could have the local TA grade a few of the papers, and then have enough overlap between them and peer graders that we can come up with reliable estimates for everyone’s grade. I’m not sure exactly what Coursera’s algorithm for that is, but I think it works pretty well.

    Working out a good model (implemented in Stan, of course) could be a neat project for the remote students.

    Final grades would be delivered in the form of samples from the posterior rather than point estimates.

  9. Sounds great, thank you.

    Video lectures, slides, and assignments is what I was hoping for. I will follow the course along with two fellow students, so we will mark each others’ assignments. Friday 8:30am Paris time works very well for us in Australia :)

  10. I have been looking for a comprehensive online course for learning Bayesian badly. I did find some scattered ones that contain some bayesian recipes, on Youtube (http://www.youtube.com/user/mathematicalmonk), Coursera (the course Probabilistic Graphic Modelling taught by Daphne Koller has some bayesian element ) and some personal websites (http://users.soe.ucsc.edu/~draper/writings.html, which has two yet finished wonderful books), but none of them are in the sense of comprehensive and data analysis oriented.

    Although reading the BDA book is a good way of learning (I do find the 5th chapter a bit confusing as well, hierarchical modelling (and prediction) might be the most powerful part of Bayesian, but it is really not easy to grasp and also not easy to teach, I guess), the PhD cohorts in our Finance and Stats department are really looking forward to taking a course like this one happening in Paris. We will follow the course when it’s on G+ hangouts.

  12. Great to know you’re coming to Paris again!Unfortunately I’ll be teaching myself on Fridays, but do you know already when your Dauphine class will be?

  13. I have been searching for a comprehensive Bayesian so badly. There are some courses with Bayesian element scattered the internet (e.g. the “mathematicalmonk’s channel” on youtube and the Coursera course “Probabilistic Graphical Modelling) but none in the sense of comprehensive and data analysis oriented.

    Thank you so much, the PhD cohort in our Finance and Stats department seems to be keen to this upcoming course. We will make the grading part a group work.
    Just wondering what is the level will this course be?

  17. Will the g+ hangout be recorded and put somewhere for those sleeping or busy at 8:30 am Paris time?

    I’m looking forward to your course but will be busy at this time. I see on the g+ website that the hangouts can be posted on youtube. It’d be so great.

Comments are closed.