This is a repost of our article, originally published in the OpenSearch blog. Metarank is an open-source secondary ranker that can perform advanced search results by reordering the results with a LambdaMART learning to rank (LTR) model. …

The 0.5.7 release of Metarank was not planned, but as we got feedback from our users about certain bugs they face, we’ve addressed them in this release. Vector Feature Extractor The vector feature extractor is able to work with arrays of numbers of variable length. The ML model can’t just work with arrays…

The latest 0.5.6 release of Metarank has not so many impressive features. However, there are a few improvements that will make your life easier. Binary format Previously, Metarank stored all the data — features, models, and click-through events in JSON format. …

“Applied machine learning is basically feature engineering” — famous quote by Andrew Ng perfectly describes why feature engineering is one of the more complicated machine learning tasks than some think. Feature engineering applies to all kinds machine learning areas — computer vision, anti fraud detection and, of course, personalized ranking…

For the past few weeks we’ve been mostly focused on fixing different bugs and issues reported by Metarank’s users, but we were able to include couple features that can improve your personalized ranking in the 0.5.5. Rate normalization Rate normalization is used to normalize values of rates. Normalization is needed to remove…

In this post we will explore how you can set up and use Metarank’s personalized ranking API with Redis persistence with Docker Compose. Data preparation Metarank is an open source personalized ranking service that you can setup quickly and enhance your user experience with personalization. From social networks to delivery services and…

Good Thursday, everyone! The Metarank team has prepared a new release with some cool features to make your life better. Auto feature We’ve added a new sub-command autofeature that can automatically create feature mapping based on your historical data. Previously, you had to come up with features yourself, creating friction in the…

Yes, we’ve skipped the announcement of 0.5.2, but we’ve described what we did there in the previous post, make sure to check it out: I know what you did with my open-source project. So, what’s new in 0.5.3? RAM Hunger After the big re-write and removal of Flink, some of the computations…

We hope that you’e had great summer and enjoyed your time off or work during the sunny days. The Metarank team has been busy analyzing the feedback that we got so far and almost re-writing Metarank from scratch. This work resulted in the new 0.5.1 version, almost a complete re-write…