What I’m working on this week

9/24/2021

I had a busy week, mostly filled with interviews and a class project. Unfortunately, “busy” doesn’t always mean “productive” on the things I’d like to be working on, but anyway, here are some of the highlights:

• I had 5 job interviews this week. I’m graduating in December, so I’m currently looking for full-time positions starting in January 2022. I tried to schedule each interview on a separate day, doing at most 2 interviews in a day, so I could put my best foot forward. Some of the interviews were technical interviews, so I needed to not be worn down and tired for the process. Interviewing takes a lot out of me… I got each of these companies from applying online, which rarely works, so I was excited to have so many companies respond.
  • Note to employers: If you make your initial technical online screening too long then you’re just self-selecting for desperate candidates who have a lot of free time. I know because I’ve been there. This semester I’m fortunate to be in a position where I can’t spend 2 hours on your online assessment. If you’re Google, you might be worth the time, but otherwise, keep your technical challenges to less than one hour, please! You can have multiple rounds of technical interviews if you really need it.
• I had a class project due this week which lasted much longer than my interest in it. It used the Amazon Fine Foods dataset to predict the number of stars for each review based on the text of the review. The difficulty of the project was in the limitations set by the professors: we could only use linear regression for prediction (which if you know anything about natural language, you know it’s not linear in any way…) which made the bulk of the project on feature engineering. We tried everything from PCA, TF-IDF, NMF and LDA topic analysis, label encoding, lexical sentiment analysis, and statistical measures like length of review and capital letters and punctuation used. Nothing worked as well as I expect TF-IDF with a simple Naïve Bayes model to work, but we were stuck with linear regression… *grumble, grumble, grumble*