Aaaaaaaaand we’re back at it again in a new year, feeling renewed, large and in charge like Sextina Aquafinaaaaaaaa bay-bayyyy~
(my favorite hyper-energetic character from BoJack Horseman. Don’t ask me why.)

(If you haven’t watched BoJack Horseman yet, you should really go ahead and do that now)
Week 07 Topics:
This week’s topics were Dimensionality Reduction, Matrix Factorizatoin & Collaborative Filtering, and Recommender Systems.
I’m gonna take a moment to go a little meta here.
I always find it kinda tough figuring out exactly what to say each week in this section:
- Do I explain the algorithms we learned? I could, but Wikipedia and many, many other resources a short google search away do that a whole lot better, and honestly, reading that stuff is pretty boring.
- Do I explain the exercises we did in class? Yea, that’s okay but I don’t wanna give any of Galvanize’s special sauce away to future students.
- Or, do I write a whole bunch of random sh*t semi-orthogonal to the topic of data science? Bingo!
Seriously though. The goal is to sound “smart”, like I really know what I’m talking about, but if I already knew it that well then I wouldn’t be enrolled in a data science program (right?).
My approach for each week’s post then, has been to just write some thoughts and general takeaways filtered through my very muddled world lens.
So to you Mr/Ms. elite-data-scientist-or-future-employer-or-galvanize-student-or-random-ass-why-the-heck-are-you-reading-my-blog-person, although I don’t talk very deeply about it, I hope you’ll be kind enough to give me the benefit of the doubt, since I have enough interest to write these posts every darn week, that maybe… just maybe… I actually do know a teeny little bit about those topics in the first sentence.
That said, huge revelation this week.
PCA/SVD (Principal Component Analysis / Singular Value Decomposition), NMF (Non-negative Matrix Factorization), and LDA (Latent Dirichlect Allocation) are all related! (dun dun duuuuun)
Before this week’s classes I honestly thought that all of these were completely different techniques and used for very different reasons (is my naivete showing yet?), turns out you can tie them together simply with a slight adjustment of the underlying assumptions and the cost function that’s used.
Mind blown.
All of these topics were geared toward teaching us about Recommender Systems which we ended the week doing a case study on.
The case study was hard.
We were given a bunch of user rating data and our job was to build the best recommender we could.
For my team, it seemed like no matter how much feature engineering ninjitsu we tried, we couldn’t get our model to improve, and our vanilla item-item similarity recommender in GraphLab ended up giving us our best score.
Only a light slap in the face.
The bottom line, shared amongst many groups, was that building a good recommender is tough.
Capstone Projects
We submitted our 3 preliminary proposals for our capstone projects this week.
As I guessed before the holiday, I couldn’t think of a respectable third project idea (not that I tried very hard), so I only submitted two:
- A product oriented idea for de-spamming the housing ads section on Craigslist and creating a web app for it
- A more exploratory idea looking at how different urban slang becomes popular
As luck would have it, I actually got feedback on my ideas while writing up this post.
Found out my first idea “although interesting” was rejected due to concerns about collecting and labeling data.
Bummer. (maybe something to do as my own personal project?)
Good news though, is that I got the go ahead to move forward with my back-up idea!
Although it falls more along the lines of a “toy” project, it’s something I’ve wanted to investigate for a while and should be a lot of fun to present!
Blog.Prettify()
I noticed the titles I’ve selected for each week’s post are very redundant and don’t tell you anything about the content of the post, which makes the main blog page look… in the wise words of my gf… “lame as hell”.

(Lame as hell on the left. Dat new new on the right.)
So I went back and changed up the titles to be a bit more descriptive and worthy of clicking on. I’d say it looks better.
Holler if you are a blogging expert and have other suggestions!
Dropping the Ball
I started this week out on a bit of a down note…
On the first day back from break, I found out the only other black student in the program dropped out.
It would be sad for any student who got into Galvanize’s Data Science Immersive and made it so far, to drop-out, but this one for me was particularly painful to hear.
I had been meeting up with him before class almost every morning to chat and prepare a little.
He had told me he was struggling, not only with the material, but also feeling at odds with some instructors and other students. In addition, he moved alone to SF for the program and lived alone, which he mentioned made him feel pretty isolated.
I tried to give him some supportive advice and was always willing to help him with questions he had. But beyond that 1 hour or so in the morning, I didn’t do anything.
I never, at the end of the day, said “hey, let’s go over that thing from earlier today..”, or “Yo, you wanna study for that assessment tomorrow?”.
And probably worst of all, I never invited him over to otherwise hangout outside of class.
The last few days before the holidays, he seemed pretty distant and I knew something was off. So when I came in on Tuesday morning and didn’t see him there, I already could sense what had happened and felt the guilt building up…
No excuses.
I of all people should know just how important having at least one person, a linchpin, can be in deciding to keep going or giving up. During the toughest parts of my life I’ve always lucked out and had at least one person to lean on.
Now I can’t ignore the voice in my head saying “So you’re telling me you’re in a program tailored specifically to introduce more underrepresented minorities into Data Science & Tech, but you just let one slip like that?… Well, nice fucking job.”
That feeling that you could’ve done more, when it’s already too late… yea, that feeling.
It sucks.
Til next week.