Time to Die 2: Improved In-Game Death Prediction in Dota

Abstract

Competitive video game playing, an activity called esports, is increasingly popular to the point that there are now many professional competitions held for a variety of games. These professional competitions are often broadcast in a professional manner similar to traditional sports broadcasts. The ability to perform moment-to-moment prediction within an esports match has numerous applications. Importantly, it has the potential to empower broadcasters to better process the oft-fast-paced nature of an esports match. This work focuses on this moment-to-moment prediction and in particular presents techniques for predicting if a player will die within a set number of seconds for the esports title Dota 2. We train our model on ‘telemetry’ data gathered directly from the game itself, and position this work as a follow up to our previous work on the challenge (Katona et al., 2019). Since the publication of that work, new dataset parsing techniques developed by the WEAVR project enable the model to track more features, namely player status effects, as well as more importantly operate in real time. Additionally, we explore two extensions to the original model, firstly we employ learnt embeddings for categorical features, e.g. which in game character a player has selected, and secondly we explicitly model the temporal element of our telemetry data using recurrent neural networks. We find that these extensions and additional features all aid the predictive power of the model. Additionally, a deeper analysis of the Time to Die model is carried out to assess it’s suitability as a broadcast aid.

Keywords: Esports, Dota 2, Deep Learning, Micro Prediction, Game Analytics, Recurrent Neural Networks