Case-based learning by observation: preliminary work

In this paper, we present an agent which uses case-based reasoning to play the real-time strategy game StarCraft. Cases are gathered through observation of human actions in particular situations, which are extracted from game log files. Cases are then used by a domain-independent case-based reasoning framework to make in-game actions based on human actions in similar situations. This work aims to demonstrate a method for more easily creating better agents in real-time strategy games.