A comprehensive resource on practically every topic for data science researchers — Edit: the compendium has its own website @ mlcompendium.com and the google doc is now fully retired in favor of the website. I’m not profiting from the compendium and I’m always trying to make sure that it is up to date, updating the body of work with new knowledge, and…

An easy approach for monitoring your ML models — We all know by now that data & model monitoring is extremely important. Even if your production-ready model didn’t change, the data distribution may have and it had potentially affected your outputs, which directly influenced your users, and once this happened you potentially have a client-facing production problem. This article…

A clever way of optimizing over metrics, models & functions — Using optimizers directly, whether minimizers or maximizers, is a subject that is rarely being discussed. These tools can be easily used to find, for example, local or global minima for any function. We’ll dive into a couple of examples the first one estimates the mode, and the second one finds…

Or why random_state=42 is an anti-pattern — The following is my opinion and reasoning about using a random seed in your work, model development & production. Regardless of the number chosen, which is in a lot of cases a well-known number, such as 42. There are certain use cases where you can use a random seed if…

A journey through nearest neighbors & its derivative algorithms — There are various algorithms that are built on the foundations of others. The following article focuses on Nearest Neighbors (NN), which other models utilize and innovate on top of their concept or code. Unsupervised Nearest Neighbors Let’s start with the foundation model, i.e., unsupervised Nearest Neighbors, implemented in Scikit’s. This algorithm is used…

How algorithms are reused in nonstandard use cases. — Academics have been known to borrow ideas from nature and other fields while applying them in a slightly different way to new problems. These days, in data science, we see many ideas, technologies, and scientific advancements that are applied across the big three (NLP, Vision, Audio). I would like to…

From the stateofmlops.com The following is an exhaustive list of companies in the world of MLOps Monitoring and their segmentation by personas, supported data type (tabular, image, audio, etc), product features (data-integrity, data-quality, health, drift, bias & fairness…

Text annotation tools are not just for creating datasets. — The very first thing that every researcher should do is to know their data and their domain. There are many ways to do it, from manually inspecting columns and rows, printing statistics, using pandas_summary, using feature selection methods, using Snorkel, visualizing data, interpreting your models, etc. I argue that the…

An insight into requirements wants and needs, and common practices. — I have had the pleasure of working with several consultants, academic experts, freelancers, or individuals that work on a per-project basis, in various situations in the last few years. The following are specific use cases for when your organization actually needs them or when someone within the organization thinks that…