존스홉킨스대학교의 Data Science Capstone 학습자 리뷰 및 피드백

one of the best

A great course!

Awesome course

Great Course

Pretty cool

it was fun

good one!

Excellent

Very Good

Excellent

engaging

Awesome.

Awesome

Thanks!

Thanks.

Thanks

Cool

Nice

Good

good

E

E

Completion of this project requires most (all?) of the skills you will have learned in completing the prerequisite courses. If you've worked to ensure you truly understand the concepts, tools and techniques presented in the prerequisite courses, you will be able to complete this project. The problem domain is a little different from most of the examples in the prerequisite courses. I find that a good thing. Whenever I learn something I believe to be useful, I always wonder how it applies in other contexts. This course was an exercise in doing just that — applying what you've learned to a "new" (i.e., new to me) a domain.

Heads up / Be aware: If you're "like me" — inexperienced with NLP, and one of those people who doesn't feel quite right about using a recommended toolset or algorithm until I understand why it's the right tool for the job — you should start reading up on the basics of text mining, NLP and next-word prediction models 1-2 weeks before you start the course. For some, that might be overkill; but I'm a slow reader at the end of a workday (we all have day jobs, right!?). Given this foundational understanding, I felt comfortable making tradeoffs among the state-of-the-art and the practical, given the project objectives, my own time constraints, etc. Reading the course forums and reviews, I think some who had trouble completing the project weren't able to take sufficient time to get oriented with this domain before attempting to build their first word prediction model.

Note: By "foundational", I mean enough to intuitively grasp why what's accepted as best practice is that. When I've read about someone's approach to solving a problem, and I'm able to say "makes sense, but I probably don't need to do X or Y to meet the need for this effort", then that's often enough… But :-) because I at times overthink things (don't we all!), I get a little more comfortable when I at least skim over descriptions of how a couple others have solved a similar problem; and I can see patterns of convergence… I do NOT mean enough to write your own thesis, unless that's what you really want to do. Whatever floats your boat! LOL

I have a software development background (and completed the previous courses in the specialization), so translating approaches I found described in various sources into code wasn't "easy"; but it wasn't a barrier, either. I was helped along GREATLY by the existence of R packages such as tm and tokenizers, and I was always able to find guidance on addressing thorny issues via "good ole Google Search". Most often, my searches would lead me to StackOverflow or write-ups from capstone project alumni. While I did my own write-ups and wrote my own code, I benefited in a big way from lessons learned by others who've already tackled similar problems.

I would recommend the Data Science Specialization by JHSU, which (as it should be) is a package deal with the capstone project. Applying what I learned to a new domain really solidified my understanding and has whet my appetite for the next challenge.

Overall, I was semi-satisfied with the capstone project:

On the negative side, my foremost issue is that the project has very little to do with what we learned in the nine courses before. I get that you will always see new data formats as a data scientist, but having the whole course cover numeric data and then having the final project be on text data where you can't apply what you learned seems sub-optimal. Also, to me it seemed that the accuracy increased mostly with how much data you train your algorithms on, and not so much how you design your algorithm. My second issue is that the class only starts every two months, and the assignments are blocked before the session starts so you can't see them if you're trying to get a head start. What happened to everyone learning at their own pace? I have a lot to do and had to switch sessions at least once for most classes, and this class was really stressful for me because I didn't want to move my completion back by two months. Lastly, I really hate RPresenter and that the instructors force us to use it, but maybe that's just me.

On the positive side, I did learn a lot: The basics of text prediction, how to do parallel programming in R and how to set up an RStudio instance on AWS (the latter two are not very hard, I recommend them to anyone struggling with gigantic runtimes, as long as you're willing to invest like $40 or so for the computing power). I liked that the guidelines were very broad, so there was a lot of room for creativity. I also finally found out how to make an pretty(-ish) presentation in R, though I would always choose Powerpoint in real life.

I really enjoyed the series as a whole and learned a great deal.

I enjoyed the course. This course took me waaaay more time than I thought because I struggled with a few issues. First, I wish I'd started by taking the NLP online course before starting the Capstone (https://www.youtube.com/watch?v=-aMYz1tMfPg). There was an issue installing RWeka, RJava and it took me several days to work through the issues. I eventually moved to using quanteda (https://cran.r-project.org/web/packages/quanteda/vignettes/quickstart.html). I also waited far too long to develop a method to test my model using a subset of the training data, so I could test whether changes to my model improved and reduced performance. It turns out that my model trained on a 25% sample performed just as well as a model trained on 100%. I'm thankful for the Discussion Forum and final peer review process. Both helped me learn how I can improve my model and demo application. I really appreciate the instructors for creating this specialization. I've learned a lot.