Doing it live: machine learning at scale... and in an instant

Ticketmaster operates in a very real-time environment, where we can go from tickets not being for sale, to being on sale, to being held for someone to purchase, to an entire event being sold out in minutes. It's a diverse environment, with well over a dozen ticketing systems, creating challenges for observing and reacting wholistically. While operations are typically consistent from one event to the next, the dynamic & evolving market means that whas was the correct market decision in one case, might not be the correct decision in the next. With sales sometimes over in minutes, waiting even an hour to react to changes in woefully inadequate. Consequently, in order to apply machine learning, we need a system & infrastructure that can consider data as it is produced from a diverse set of systems and learn new decisioning logic as it receives each new piece of data.

This talk will describe how Ticketmaster has combined open source tools like 0mq, Kafka, Storm, Vowpal Wabbit, etc. that allows millisecond reactions to a highly dynamic ecosystem and market place, allowing systems to learn in real-time to respond to & learn from market conditions that might not have existed moments before. It will also discuss how we leverage data tools like CKAN to provide accurate and up to date catalog of data across a diverse data ecosystem.