Cooperative Learning by Replay Files in Real-Time Strategy Game

Kim, Jaekwang; Yoon, Kwang Ho; Yoon, Taebok; Lee, Jee-Hyong

doi:10.1007/978-3-642-16066-0_7

Cited by 8 publications

(3 citation statements)

References 1 publication

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The datasets of this group tend to be compact, but their use is limited to build order prediction. Both (Hsieh and Sun 2008), (Kim et al 2010), and (Synnaeve and Bessière 2011a; Synnaeve and Bessiere 2011b) have done build order prediction as a task.…”

Section: Related Workmentioning

confidence: 99%

STARDATA: A StarCraft AI Research Dataset

Lin

Gehring

Khalidov

et al. 2021

AIIDE

View full text Add to dashboard Cite

We release a dataset of 65646 StarCraft replays that contains 1535 million frames and 496 million player actions. We provide full game state data along with the original replays that can be viewed in StarCraft. The game state data was recorded every 3 frames which ensures suitability for a wide variety of machine learning tasks such as strategy classification, inverse reinforcement learning, imitation learning, forward modeling, partial information extraction, and others. We use TorchCraft to extract and store the data, which standardizes the data format for both reading from replays and reading directly from the game. Furthermore, the data can be used on different operating systems and platforms. The dataset contains valid, non-corrupted replays only and its quality and diversity was ensured by a number of heuristics. We illustrate the diversity of the data with various statistics and provide examples of tasks that benefit from the dataset.

show abstract

Section: Related Workmentioning

confidence: 99%

STARDATA: A StarCraft AI Research Dataset

Lin

Gehring

Khalidov

et al. 2021

AIIDE

View full text Add to dashboard Cite

show abstract

“…Weber et al encode the whole game into a vector for each player and then model the problem into a supervised learning problem (Weber and Mateas 2009). Kim et al propose to categorize different building orders and summary them into separate states and actions (Kim et al 2010). Different from the above work, Hostetler et al start to consider partially observe problem in RTS game by using dynamic Bayes model inference (Hostetler et al 2012).…”

Section: Related Workmentioning

confidence: 99%

Macro Action Selection with Deep Reinforcement Learning in StarCraft

Kuang

Zhuang

et al. 2019

AIIDE

View full text Add to dashboard Cite

StarCraft (SC) is one of the most popular and successful Real Time Strategy (RTS) games. In recent years, SC is also widely accepted as a challenging testbed for AI research because of its enormous state space, partially observed information, multi-agent collaboration, and so on. With the help of annual AIIDE and CIG competitions, a growing number of SC bots are proposed and continuously improved. However, a large gap remains between the top-level bot and the professional human player. One vital reason is that current SC bots mainly rely on predefined rules to select macro actions during their games. These rules are not scalable and efficient enough to cope with the enormous yet partially observed state space in the game. In this paper, we propose a deep reinforcement learning (DRL) framework to improve the selection of macro actions. Our framework is based on the combination of the Ape-X DQN and the Long-Short-Term-Memory (LSTM). We use this framework to build our bot, named as LastOrder. Our evaluation, based on training against all bots from the AIIDE 2017 StarCraft AI competition set, shows that LastOrder achieves an 83% winning rate, outperforming 26 bots in total 28 entrants.

show abstract

“…AI researchers analyze the replays to build a "strategy" prediction model using CBR [9], J48, k-NN, NNge [2] and Bayesian network [10]. Also, it helps to find the goal of players [11] and relationships among build-orders (strong or weak) [12].…”

Section: Starcraft Replay Miningmentioning

confidence: 99%

Replay-based strategy prediction and build order adaptation for StarCraft AI bots

Cho

Kim

Cho

2013

2013 IEEE Conference on Computational Inteligence in Games (CIG)

View full text Add to dashboard Cite

StarCraft is a real-time strategy (RTS) game and the choice of strategy has big impact on the final results of the game. For human players, the most important thing in the game is to select the strategy in the early stage of the game. Also, it is important to recognize the opponent's strategy as quickly as possible. Because of the "fog-of-war" in the game, the player should send a scouting unit to opponent's hidden territory and the player predicts the types of strategy from the partially observed information. Usually, expert players are familiar with the relationships between two build orders and they can change the current build order if his choice is not strong to the opponent's strategy. However, players in AI competitions show quite different behaviors compared to the human leagues. For example, they usually have a pre-selected build order and rarely change their order during the game. In fact, the computer players have little interest in recognizing opponent's strategy and scouting units are used in a limited manner. The reason is that the implementation of scouting behavior and the change of build order from the scouting vision is not a trivial problem. In this paper, we propose to use replays to predict the strategy of players and make decision on the change of build orders. Experimental results on the public replay files show that the proposed method predicts opponent's strategy accurately and increases the chance of winning in the game.

show abstract

Cooperative Learning by Replay Files in Real-Time Strategy Game

Cited by 8 publications

References 1 publication

STARDATA: A StarCraft AI Research Dataset

STARDATA: A StarCraft AI Research Dataset

Macro Action Selection with Deep Reinforcement Learning in StarCraft

Replay-based strategy prediction and build order adaptation for StarCraft AI bots

Contact Info

Product

Resources

About