Jeez, week 04 already. Hard to believe the first month at Galvanize is already over.
At this point my data science all day err’day schedule is now just an afterthought…
Week 04 Topics:
This week we dug even deeper into the machine learning world, learning about Cross Validation, Regularization, Logistic Regression, Gradient Descent, Imbalanced Classes and Profit Curves.
Luckily, I’m very familiar with those topics thanks in large part to the phenomenal Python Machine Learning by Sebastian Raschka. If I ever meet Sebastian in real life I’ll give him a big ol’ hug and buy him a couple boxes of Jamaican beef patties, that’s how much I love his book.
I had tried watching classes on machine learning, reading other books, and a few other things to learn, but although I picked up little tidbits here and there it never really worked. With PyML though, the way Sebastian explained everything and then laid it out in code, it finally all “clicked” for me in a big way.
It has always been interesting to me why resources that get great reviews from many other people might fail to work for you, then something else comes along and shows you the light… like voodoo.
PyML was my gateway to everything machine learning, and I go back to it frequently whenever I need to review, so I’m sure I’ll be going to it often during the rest of the Data Science Immersive (DSI).
That said, the individual and pair exercises that Galvanize instructors have set up were great practice and really test if you know the topics as well as you think you do, and if you can translate that into code?
I found implementing oversampling & undersampling tougher than expected even though the general idea is really simple. A good discovery to make.
One topic that I had never come across before were profit curves, which is calculating the profits and costs a model will generate from it’s confusion matrix. I think it’s a pretty nifty idea since it allows you to give a concrete answer to the most important business question ever:
“How much guap we gon’ make?”
Holiday Party:
On Thursday Galvanize organized a holiday party for alumni and current students. It was a nice break from the normal daily grind, and was fun to join other students and let loose for once.
We were also allowed to bring a +1 to the party with us, so I brought my girlfriend with me to give her a tour of the campus and give her a look behind the scenes at the place that’s been occupying the majority of the time she used to get. Hopefully it’s a bit of consolation.
Thanks for your endless understanding and support! @Grace
Additionally, I finally discovered the other Airbnb Data Science Associate in my cohort!
We talked about the DSI and our thoughts on moving on to Airbnb afterwards. It wass cool to be able to trade thoughts since our experience for the next 5+ months will be so similar.
On another note, the few other Airbnb Data Science Associate scholarship winners I’ve met have told me they are very reluctant to reveal that they got the scholarship.
I can understand where this is coming from. Fear of getting a target placed on your back, getting labeled a product of Affirmative Action, being judged, extra pressure, etc. But I believe this is a limiting belief and sadly also kind of does disservice to the reason why Airbnb and other companies have had to resort to creating programs like it in the first place.
For me, I’ve been very open about the scholarship and every student, instructor, random techist (and anyone reading this blog) who I’ve told about the program has been extremely positive and genuinely wanted to know more about the issues involved. Obviously, they could all be faking it, outwardly positive and inward haters, but I tend to believe humans, in general, are good.
So much more to say about this, but I’ll again leave it at this for now (…I really should get on writing that post about minorities in tech huh?)
Regression Case Study:
On Friday we went through our first case study, getting into groups of 4 and trying to use everything we’ve learned up til now to make a model for predicting heavy machinery prices. We then presented our results to other groups and instructors at the end of the day.
It was exciting and fun competing with the other teams, even though we lost…
But the major lesson of the day; it doesn’t matter how well you know ML, operational and communication problems within your team plus an utter lack of domain knowledge will absolutely kick your ass!
My team had some really great ideas, that were confirmed by instructors at the end of the day, but we failed to implement them due to time constraints and the problems listed above.
This was definitely the most important lesson of this week, and is something to keep in mind for the next case study and when working in industry side-by-side with a team of other people.
Sharing is Caring:
I’ve been meaning to gather links to the websites and blogs of other students in my cohort(19/20) who are also writing about their experience at Galvanize.
Reading previous DSI student’s blogs was a lot of help getting an idea of what to expect in the program, so I imagine it’ll be doubly helpful for future students to read other perspectives on the program from people going through it at the same time.
So first up is Jenny over at pltalot, who has a great collection of analysis projects and is also writing about her thoughts on the DSI. So go check her blog out!
I’ll go back and add all the links I get to my first and final posts on Galvanize, which I’m guessing will be the ones most prospective students will read.
Til next week!