はてなブックマーク

Choosing a Sequential Testing Framework — Comparisons and Discussions March 21, 2023 Published by Mårten Schultzberg (Staff Data Scientist) and Sebastian Ankargren (Sr. Data Scientist) TL;DR Sequential tests are the bread and butter for any company conducting online experiments. The literature on sequential testing has developed quickly over the last 10 years, and it’s not always easy to determine

February 1, 2023 Published by Divita Vohra, Sr. Product Manager, Keshi Dai, Sr. ML Engineer, David Xia, Sr. ML Engineer, & Praveen Ravichandran, Staff Research Scientist As the field of machine learning (ML) continues to evolve and its impact on society and various aspects of our lives grows, it is becoming increasingly important for practitioners and innovators to consider a broader range of pers

March 17, 2022 Published by Alexandre Tamborrino, Machine Learning Engineer Beyond term-based Search Until recently, Search at Spotify relied mostly on term matching. For example, if you type the query “electric cars climate impact”, Elasticsearch will return search results that contain everything that has each of those query words in its indexed metadata (like in the title of a podcast episode).

The Rise (and Lessons Learned) of ML Models to Personalize Content on Home (Part I) At Spotify, our goal is to connect listeners with creators, and one way we do that is by recommending quality music and podcasts on the Home page. In this two-part blog series, we will talk about the ML models we build and use to recommend diverse and fulfilling content to our listeners, and the lessons we’ve learn

Introducing Pedalboard: Spotify’s Audio Effects Library for Python September 7, 2021 Published by Peter Sobot, Staff Machine Learning Engineer - Spotify Audio Intelligence Lab We’ve just open sourced Pedalboard, Spotify’s framework for adding effects to audio in Python. Pedalboard makes it easy to use studio-quality audio effects in your code, rather than just in your digital audio workstation (DA

It’s All Just Wiggly Air: Building Infrastructure to Support Audio Research TL;DR We just open sourced Klio — our framework for building smarter data pipelines for audio and other media processing. Based on Python and Apache Beam, Klio helps our teams process Spotify’s massive catalog of music and podcasts, faster and more efficiently. We think Klio’s ease of use — and its ability to let anyone le

Yesterday, we released the open source version of Backstage, our homegrown developer portal. And we learned a thing or two via the feedback we received. So, I wanted to take this opportunity to further explain what we’re trying to do with Backstage — and more importantly, what we want to give to the greater engineering community beyond Spotify. What’s the big infrastructure problem? As companies g

The Winding Road to Better Machine Learning Infrastructure Through Tensorflow Extended and Kubeflow When Spotify launched in 2008 in Sweden, and in 2011 in the United States, people were amazed that they could access almost the world’s entire music catalog instantaneously. The experience felt like magic and as a result, music aficionados dug in and organized that content into millions of unique pl

August 31, 2018 Published by Erik Carlsson, Eirini Kakogianni We flipped one server flag and got more download bandwidth for Spotify users. That is the TL;DR of this A/B experiment with BBR, a new TCP option. Background BBR is a TCP congestion control algorithm developed by Google. It aims to make Internet data transfers faster, which is no small feat! How Spotify Streams Music The basic principle

For more recent information on Spotify’s Health Check model, check out this blog post. What is a squad health check model? A lot of companies experiment with ways of measuring and visualizing how their teams are doing. They’re usually called “maturity models”, and involve some sort of progression through different levels. The intent of these types of models is usually benign – for example managers

Why do we write tests? Most people would say that we write tests to verify that things work as we expect them to. While that is true, it’s not the whole truth. After all, that can be verified through manual tests as well. So there has to be something more to it. Anyone who has ever done manual testing knows that it’s slow, boring and error-prone. By writing automated tests we are trying to remove

TL;DR: We have created the Retro Kit. It can be downloaded by following this link. Retro Kit… but why? Historically at Spotify there has been an Agile Coach attached to each team. That coach has, among other things, helped the team to improve their processes. A central piece to that improvement are regular retrospectives where teams reflect on their way of working to continuously become better in

Spotify’s Event Delivery system is responsible for delivering hundreds of billions of events every day. Most of the events are generated as a response to a user action, such as playing a song, following an artist or clicking on an ad. All in all, more than 300 different types of events are being collected from Spotify clients. The Event Delivery system is one of the core pillars of Spotify’s data

In this part we’ll take a closer look at Scio, including basic concepts, its unique features, and concrete use cases here at Spotify. Basic Concepts Scio is a Scala API for Apache Beam and Google Cloud Dataflow. It was designed as a thin wrapper on top of Beam’s Java SDK, while offering an easy way to build data pipelines in idiomatic Scala style. We drew most of our inspiration for the API from S

Whenever a user performs an action in the Spotify client—such as listening to a song or searching for an artist—a small piece of information, an event, is sent to our servers. Event delivery, the process of making sure that all events gets transported safely from clients all over the world to our central processing system, is an interesting problem. In this series of blog posts, we are going to lo

Introduction When you log into Spotify, browse through your Discover Weekly playlist, and play a track, you’re interacting with some of our fleet of around 12,000 servers. Spotify has historically opted to run our core infrastructure on our own private fleet of physical servers (aka machines) rather than leveraging a public cloud such as Amazon Web Services (AWS). Our fleet consists of a minimal s

Whenever a user performs an action in the Spotify client—such as listening to a song or searching for an artist—a small piece of information, an event, is sent to our servers. Event delivery, the process of making sure that all events gets transported safely from clients all over the world to our central processing system, is an interesting problem. In this series of blog posts, we are going to lo

“You just talk to them for half an hour.” That’s the guideline I got when I first joined Spotify for how to run my 1:1s — a recurring half hour meeting between a manager and their team members. It didn’t leave me with much to work with. I was admittedly clueless about the practice. I spent ten years in an entirely different culture, where managers managed people and teams, assigning tasks, trackin

All of our lovely Spotify users generate many terabytes of data every day. All the songs that are listened to, all the playlists you make, all the people you follow, and all the music you share. Somehow we need to organise, process and aggregate all of this into meaningful information out the other side. Here are just a few of the things we need to get out of the data: Reporting to record labels a