Moving from academia to industry — If you are a graduate student in a quantitative discipline (engineering, biology, even sociology), you can make it in industry as a data scientist. But you do need a bit of preparation because data science is more than just statistics. How to prepare for the interview You have the necessary mathematics and programming skills, but are…

Why a simple question is so effective at making you work smarter — You hear the term “10x” bandied about so much that its lost its purpose, which is to shock. It does not mean you need to work harder (grit or “not giving up”) or become more skilled (“10x engineers”). Instead, 10x is about working smarter, by changing your approach to a…

How to adapt custom Layers, Model, loss, preprocessing, postprocessing into a servable API — If the only Keras models you write are sequential or functional models with pre-built layers like Dense and Conv2D, you can ignore this article. But at some point in your ML career, you will find that you are subclassing a Layer or a Model. Or writing your own loss function…

The obligatory manifesto that one needs to publish when leaving Google — I have always believed in “three strikes and you are out”. And that is why I decided to leave Google. First, we had this great idea to start Google Wind. But Google killed the project by not being willing to scale this beyond the Netherlands. Not enough windmills, they said.

How it works, and what you can do with it — BigQuery offers the ability to invoke user-defined functions (UDFs), but those UDFs have to be in SQL or JavaScript. Such functions run entirely in BigQuery and, therefore, offer great scalability and performance (SQL more so than JavaScript). But there are lots of situations where you need to run code outside…

Drive transformational improvement in users’ workflows, not an incremental improvement in the tools you use — Many years ago, my grandmother asked me what my job was. I explained to her that I did weather research. “But what is it that you make?,” she asked. We went around and around and I don’t think I ever gave her a satisfactory answer — she would have understood…

In the majority of cases, a feature store is overkill — It appears that every sophisticated ML team has built a feature store for their ML platform. Uber built Palette. Airbnb built Zipline. Netflix built Time Travel. Google Cloud worked with our customer GoJek to build Feast. Fortunately, you no longer need to build or manage your own. Google Cloud Vertex…

Linear programming on streaming data within an Apache Beam pipeline — Not everything needs to be machine learning! Classical operations research methods can be quite good for optimizing processes, product-mixes, and logistics if you understand the constraints you are operating under. You can even operate on streaming data within an Apache Beam pipeline to achieve real-time control and agility. Problem