In Case You Missed It

2019 has been the year of the blog so far for me. I started blogging seriously in May and five months later, I must say that it’s been one of the more rewarding professional endeavors I have ever taken on. Through my blog, I’ve learned new things, met new people, and developed a more disciplined and diligent approach to both work and life.

I am happy to report that I have now written enough blogs to justify a retrospective. So today I will highlight five blog posts of mine – these are the ones that I found most enjoyable to learn and write about so far. Without further ado, here is the list.

[Understanding Neural Networks](https://towardsdatascience.com/understanding-neural-networks-19020b758230) – I worked my butt off on this piece. To this day, it remains the post that I spent the most time researching, writing, and creating graphics for. It wasn’t my most read or clapped piece but it is the one that I am proudest of. Neural nets are what most data folks currently think of when they think artificial intelligence. In this post, I do my best to explain how neural nets work in simple English and with friendly visuals. I learned a ton writing this piece and still refer back to it from time to time when I need to work with neural nets (I am forgetful). Out of all my posts, Understanding Neural Networks best represents what this blog is trying to be about.
A Better Way to Skill Up – While it is one of my least read pieces to date, it was also one of the most fun to write. I am really passionate about the bootcamp industry (and Education tech)— I recently completed Metis’ Bootcamp, loved it, and I am chock full of (unsolicited) opinions about how to improve the experience. The biggest drawback to education tech currently is what happens afterwards – as fun as learning is, we still need to find a decent job so we can pay the bills. And while bootcamps et al. do a great job of teaching students the fundamentals, they currently don’t do enough to help their students establish the necessary credibility to get hired (this problem is especially stark for career changers). In this post, I dream up a business idea that might address this shortcoming.
Do I Have Enough Money to Retire? – Developing and running Monte Carlo simulations of investors’ wealth leading up to and throughout retirement was a big part of my job for the past few years. Working behind the scenes, I never got to directly feel the impact that my company’s products had on its customers. So it was really gratifying to share some of the knowledge I gained on the job via this post – and if it helps just a few people improve their retirement readiness then awesome!
[The Curse of Dimensionality](https://towardsdatascience.com/the-curse-of-dimensionality-50dc6e49aa1e) and [Understanding PCA](https://towardsdatascience.com/understanding-pca-fae3e243731d) – I cheated and picked two. Because these two posts are intimately related with each other. The Curse of Dimensionality explores why high dimensional data can be troubling to data scientists. I personally got a lot out of Writing this post – because prior to writing it, while I knew high dimensional data was troublesome, I didn’t really know why. Doing the research and coming up with the examples and illustrations really cemented the concept for me. Understanding PCA explores principal components analysis, a popular algorithm for dealing with high dimensional data. PCA is an immensely useful algorithm. You’ve probably heard of topic modeling in natural language processing right? Well PCA is to numbers what topic modeling with LDA or NMF is to words – PCA takes in your features and extracts the key latent trends (a.k.a. topics) in your data. These latent trends help improve your understanding of the data and ultimately allow you to build better models.
Let’s Play Blackjack (with Python) and Teaching A Neural Net To Play Blackjack – once more I cheated and chose two posts. Coding up and writing these two took a lot of time – it really felt like I was back in the bootcamp again. It’s not easy writing more involved pieces like these now that I am working full time. But the feeling of achievement I get when I finally hit publish is a pretty great one. The neural net post was especially interesting. At first I thought it would be trivial to put together – just generate data with my blackjack code, drop it into a neural net, and voila. But it was anything but – I actually spent a fair bit of time thinking about the best way to simulate my target variable so that the output of a model trained on it would be useful. This project was a great reminder that while everyone loves models, the real work of a data scientist (and what differentiates a good analyst/scientist from a great one) is developing a deep understanding of your data and the questions that you are trying to answer with that data.