notes on designing ML systems.
Table of contents The product page A naive (yet effective) algorithm. Product To Vectors Product Graphs Putting it all together - Personalization, Utility and Rules Pitfalls in Evaluation Closing thoughts Further Reading After the introductory overview of e-commerce recommenders, it’s now time to get to the Related Products strategy. The biggest chunk of revenue that flows th...
Buying stuff without actually stepping out is one of those things that has radically changed all our lives. In this post , I want to dig deeper into how recommendations power online shopping. There is a lot of machine learning and also a fair amount of business practicalities and seeing how these systems are designed can teach us a lot. Some context In the bad old days of the internet, e-comme...
Last time we looked at how to best utilize a GPU when training Deep Learning models. Now it is time to put that to use and build a model that can tell apart the bewilderingly similar looking species of flora and fauna. This kind of image classification is generally called Fine grained visual categorization and take a look at these two very similar looking but different species of dragonflies to...
The last couple of weeks, I have been playing around with data from my favorite app, iNaturalist. The good folks there, have released a dataset with 2.6M images of 10,000 species of assorted beasts and other flora. I’ve been trying to build a model that can separate the Jungle Babbler from the Yellow-Billed Babbler, a task that would be beyond most non-ornithologically minded humans. Table of ...
Welcome to the fourth post in the How Ads work series. Now it’s time to earn some money and sell ads ! This will be a short post and my goal in this post is to outline the basic mechanics of how the ads auction runs and complete our end to end understanding of the entire ads funnel. Table of contents Why do we need an Auction First pass at a solution Generalized Second Price Auctions ...