After an extremely long wait, today was the day that the fifth course in Coursera’s Machine Learning Specialization was set to begin. I’ve been with this specialization since it launched in the fall of 2015. Students were initially promised an ambitious slate of six courses, including a capstone that would wrap up by early summer of 2016. With noted husband and wife couple Carlos Guestrin and Emily Fox, previously of Carnegie Mellon and now of the University of Washington, this sounded like a great option. I came to data science in large part thanks to the Johns Hopkins Coursera Data Science Specialization. However, those courses emphasized R over Python and data wrangling over machine learning. Here was the opportunity to develop additional knowledge in another direction.
Unfortunately, the specialization had problems from the beginning. On the one hand, I’ve reviewed all the classes in the machine learning specialization and found them to be quite good. Though the quality of the instructional videos from Guestrin and Fox were outstanding and the assignments did a good job reinforcing those concepts, there were other issues. First, many students came away frustrated that the courses emphasized the use of Graphlab Create, a closed source software from Guestrin’s company Turi (formerly Dato) over open source libraries like Scikit-Learn. It’s true that Graphlab was not a requirement to complete the coursework, but using an open source solution such as Scikit-Learn added a significant level of commitment given that the lessons had been tailored with starter code to Graphlab.
Second, and probably more frustratingly, the original timeline that had been announced for the classes was plagued with delays and never came close to materializing. While there was supposed to be a couple of weeks between classes, it would balloon to several months. Timelines were announced, and then delayed with new timelines, and then delayed again. When the sale of Turi to Apple for $200 million was announced, many students came to believe that the real reason behind the delays had been a focus on the sale of the company. Many were concerned about whether the specialization would even continue and if they’d be able to continue to use Graphlab, now owned by Apple. However, a statement was issued by Turi (approved by Apple) promising that Graphlab would continue to be available for academic use and the specialization would continue.
Today, as the fifth class was set to begin, a class many had eagerly anticipated and planned their professional development calendar around, the digital doors did not open. No course materials were available. No class forums were to be found. I reached out to Coursera support and got this response.
Thanks for contacting us.
Due to unforeseen circumstances, this course will now be launching as a stand-alone course (not part of the Specialization) sometime around the month of May. We know you’ve been patiently waiting for this course to become available for many months and we’re terribly sorry to be the bearer of such unfortunate news. We appreciate your understanding that this situation is completely out of our control and that our team is taking every measure we can to make this a good experience for you and all Machine Learning Specialization learners.
The Specialization is being slimmed down to the first four courses only. Once you’ve earned Certificates in courses 1 – 4 Coursera will issue your Specialization Certificate. This may take up to a few weeks as we build this process out, we appreciate your patience in the meantime.
The instructors are still committed to launching Recommender Systems and Dimensionality Reduction and hope to do so at the end of April/early May. Both the University of Washington and Coursera believe that it’s in everyone’s best interest at this point to launch the course separate from the Specialization so that learners are no longer kept waiting to complete their Specialization. If you pre-paid for the Specialization you will not need to submit additional payment to gain access to this course. There will not be a Capstone Project associated with the four course Specialization.
In this case I think the best option is to give you the refund for your payment and if you want you can take the course when is launched as a individual one.
We hope to hear your answer as soon as possible. You have until Feb 12 to request the refund.
Coursera Community Support
In summary, the original six course specialization I signed up for has been abbreviated to four, with an additional class cobbled on, best case a year after the fourth course was launched. I see that in the time I’ve been working on this post the Coursera site has quickly been changed: all references to the recommender class have been removed, and the specialization now only refers to six classes, not four (Turi’s site still refers to the six courses as of this post).
I don’t want to be overly dramatic about the result of this sequence. Coursera has promised a refund, which of course they should. I’m not here to rant. I am here, however, to point out that Coursera has changed noticeably in its 5-year existence. Not all of those changes are for the benefit of students, at least not to the benefit of all students. This is a for-profit venture, and that has really begun to show in the last couple of years.
I started taking Coursera classes in 2014, and I probably got in on the tail end of the era where it felt like Coursera was about democratizing education across the world. It felt like the pitch was, “Anyone in the world can learn from the best professors on earth regardless of their means.” Today it feels like the pitch is, “Advance your career in business or technology for under $100 per class.” Whereas it was once “Come one, come all,” there are now significant restriction of course materials without payment, and the emphasis seems to be on the types of classes (i.e. tech) where students will pay for skills that will lead to raises. I enjoy the tech classes myself and have benefited tremendously from them, but they are not the whole of education. A current glance at the Coursera landing page, and you have to scroll all the way to the bottom, past row after row of data science, programming, and business courses, before you see a mention of social sciences. No doubt, this is an emphasis on the most profitable courses. As an aside, I’m currently taking “The Science of Happiness” on EdX and finding it to be an outstanding class despite the fact that it won’t advance my career one bit.
In this environment, I think it’s unsurprising when we see an example where course development took a backseat to company development, or where a startup’s software is featured, or where teachers are now selling ebooks to accompany the classes where they once found free supplemental resources (as I have seen and heard of in other classes). I’m not saying these things are necessarily wrong. If I had a company I could sell for $200 million, you’d better believe I’d put some other priorities on hold for a while.
But as Coursera changes their expectations for their students, they’d better be prepared for a different set of expectations from their students. If you are asking people to fork over nearly $100 every time out, and you aren’t able to tell me when or if that class is going to start, you’re going to start losing some repeat buyers. It will certainly have me vetting future classes with a bit more caution before I part dive in as these specializations are an admittedly reasonable investment of money, but more importantly, of the one thing we can never be refunded — time.
- Coursera Review: Machine Learning Foundations—A Case Study Approach
- Coursera Review–Machine Learning: Regression
- Practical Machine Learning Coursera Review