Part 2: BI for decision-making This is part 2 of this series. You can find part 1 here. There are situations where you need purely historical measurements, sometimes of standard metrics that can be used to compare to other companies, like with investors or management boards or sometimes management. But…

Part 1: Data Doesn’t Lie…Well, Sort of. This will be a multi-post series. Later posts will get into more specifics but here I will just describe current issues and a high-level vision of “data as a partner”. A lot of companies say that they are data-driven. Although I don’t think…

I recently read https://medium.com/towards-data-science/do-you-really-need-a-feature-store-e59e3cc666d3, and although it is good, there are a few more things to think about (at least these weren’t apparent to me, but maybe they are mentioned in someway there) Calculating features takes longer than you want You will probably have to pre-calculate and store values, if…

Measuring how well a model’s predictions are is, of course, crucial since you want to replace what you already have only with something better (or introduce something new but choose the right path for it). But comparing the performance of different models, so that you can be confident about which…

This paper was intriguing https://arxiv.org/abs/2003.08505 in part because finding similar items is such a big part of many recommendation systems or similar systems (e.g. https://medium.com/airbnb-engineering/listing-embeddings-for-similar-listing-recommendations-and-real-time-personalization-in-search-601172f7603e and https://ai.facebook.com/blog/powered-by-ai-instagrams-explore-recommender-system/). But this paper reminded me of a few other things.

Given the interviews I’ve experienced, discussions of AI or other automated methods taking over data science, etc. I worry there’s a common misconception about technical expertise. The way I see it, there are 3 types to technical expertise: Technical knowledge. More or less, given a specific task, complete that task…

I’ve experienced quite a few technical interviews and noticed that most are bad (I mean they don’t give an indication of whether the candidate can do the job being hired for), at least for more senior positions. I’m obviously not the only person with this opinion (see, for example, https://theoutline.com/post/1166/programmers-are-confessing-their-coding-sins-to-protest-a-broken-job-interview-process?zd=2&zi=knyxnghe…

First, let me specify a few caveats: This is not meant as a comparison, so other products may be better or worse. This is about a fairly specific type of data science that is analysis and prototyping heavy. …

In my experience, people often want some metric to inform their decisions, but often the only metrics they think to ask for (probably because they are the only ones they know) are industry norms, but do not actually directly answer the questions the people have. I am a big advocate…