Keeping Up With Data — Week 13 Reading List
Just yesterday I read an article about why not to start a data science consulting business by SeattleDataGuy. Given the fact that’s what I’ve been doing for the most of the last six years the advice came a bit late! One of the reasons — “you may think you’re going to be your own boss — but you’ll actually be working for more people than before” — made me laugh at first. Maybe because it’s true. But at the same time, I find the situation with multiple ‘bosses’ strangely liberating since it comes with variety and optionality.
This week’s list is slightly tilted towards machine learning, but there is some variety in it too.
- How We Built A Context-Specific Bidding System for Etsy Ads: Etsy enables sellers to promote their items through Etsy Ads and now they’ve shared how they built the auto-bidding system to make it easier for them. I’m a big fan of the Etsy’s data blog — Code as Craft. They always start with a business problem (automated bidding), lay down the ground work (second-price auction and importance context-specific parameters), cover the machine learning meat (neural network architecture for a contextual post-click conversion rate prediction model and learning-to-rank methodology) and finish with evaluation (the buyers making more purchases from ads and the sellers receiving more sales for every dollar they spent on Etsy Ads). (Code as Craft)
- Data Scientist vs Machine Learning Engineer Skills. Here’s the Difference. There are a lot of skills needed to build an end-to-end, production-ready data-powered solution. So, it’s no wonder that companies are sometimes looking for multiple roles to get the job done. Data scientists and ML engineers are among the most popular roles. This article nicely summarises the skills needed as well as how the roles supplement one another. I’d argue that the role described here as ML Engineer is actually a Data Engineer, but I do see that a position with “ML” in the title is more attractive to potential candidates. (Matt Przybyla @ Towards Data Science)
- Trending Toward Concept Building — A Review of Model Interpretability for Deep Neural Networks: Explainability is one of the principles of ethical AI. It’s natural to ask the ‘why’ question. Even if it is targeted at a deep neural network. One way is to look at the importance of individual input features — such as pixels, in case of images. Then come concepts either simple (like corners and edges) or complex (like a face of a cat). Now, there are methods focusing not only on extracting the concepts important for an existing model but building a model with the ambition for its higher-level concepts to be human-readable. And there is a chance that optimising for interpretability will lead to better prediction accuracy too. (Domino Data Lab)
I remember watching a documentary about the Czech pickup artist community many years back. I found it bizarre. Probably because I married my high school sweetheart, and my only pick-up line was “do you want to go out with me?” But this week it got even weirder — now, there are GPT-3-generated pick-up lines. Glad I’m not on the dating scene!