|
|
Using reinforcement learning to learn how to play text-based games
Zelinka, Mikuláš ; Kadlec, Rudolf (advisor)
The ability to learn optimal control policies in systems where action space is defined by sentences in natural language would allow many interesting real-world applications such as automatic optimisation of dialogue systems. Text-based games with multiple endings and rewards are a promising platform for this task, since their feedback allows us to employ reinforcement learning techniques to jointly learn text representations and control policies. We present a general text game playing agent, testing its generalisation and transfer learning performance and showing its ability to play multiple games at once. We also present pyfiction, an open-source library for universal access to different text games that could, together with our agent that implements its interface, serve as a baseline for future research.
|
|
|
Practical Exercises for Artificial Intelligence
Mařík, Matouš ; Švancara, Jiří (advisor) ; Zelinka, Mikuláš (referee)
The thesis describes practical tasks for the lecture of Artificial Intelligence, that is to introduce students to basic structures and algorithms of this field of informatics: rational agent, uninformed and informed search, adversarial search and algorithms solving constraint satisfaction problems. It consists of the motivation for their creation, their implementation and documentation.
|
|
|
Artificial intelligence for Patchwork board game
Chudoba, Michal ; Holan, Tomáš (advisor) ; Zelinka, Mikuláš (referee)
Bachelor thesis on the topic of artificial intelligence in the game Patchwork. It deals with the actual implementation of the game and the computer-controlled player, as well as approach that was taken to get a com- petent opponent. The developed program allows the user to play against different types of opponent, but it also gives a demonstration of simulation between pla- yers. An analysis was made as to whether it was advantageous or disadvantageous to start the game.
|
|
|
General Chess-board Games
Emmerová, Petra ; Pilát, Martin (advisor) ; Zelinka, Mikuláš (referee)
This work deals with games played on chessboard and examines which algorithm for playing them performs the best. First, an application capable of playing chessboard games, and using a user interface or text file to define your own game composed of elements of these three games was implemented. Furthermore, a minimax algorithm with alpha-beta pruning and Monte Carlo Tree Search was implemented so that it would be able to play all user-created games. Subsequent comparisons then showed that the minimax algorithm with alpha-beta pruning has the best results for general games on chessboard. 1
|
|
|
Artificial Intelligence for the Game Fantasy Realms
Miklóšová, Tereza ; Hric, Jan (advisor) ; Zelinka, Mikuláš (referee)
In this thesis we work on the implementation of the card game Fantasy Realms for simultaneous play of more than one player and the artificial intelligence capable of playing this game. The artificial intelligence is based on greedy algorithm which we reworked to greedy algorithm with greater insight and on reinforcement learning. Greedy algorithm has been proved to be a good model for playing this game and has achieved great results in regard to average score. On the other hand learning agent based on reinforcement learning has not been very successful, because the model of the game we provided to it was not satisfactory for learning purposes. Thus the agent learned by reinforcement learning could not overcome the greedy one. 1
|
|
|
Artificial intelligence for the Game Carcassonne: The Discovery
Motlíček, Ondřej ; Hric, Jan (advisor) ; Zelinka, Mikuláš (referee)
The bachelor paper deals with the development of an artificial intelligence for the game Carcassonne - The Discovery. Different approaches for designing an artificial intel- ligence are presented. Heuristic functions based on various aspects of the game. Monte Carlo methods and the Expectiminimax algorithm are used for state space of the game. The designed methods are implemented and experimentally tested and compared by simulations of the game between the artificial players. Results of the experiment are pre- sented and explained. The simulation environment consists of multiple programs for the game simulation of both artificial and human players. A batch simulation of the artificial intelligence is emphasized. 1
|
|
|
Simplified Versions of Chess for the Teaching of the Game
Hübsch, Anna ; Pilát, Martin (advisor) ; Zelinka, Mikuláš (referee)
Despite its age, chess is still one of the most popular and most played board games. Nowadays, it is more and more common to train even little children and simplified versions of chess are often used for that. The pur- pose of this thesis is to create an application for training of children and to compare various methods of artificial intelligence. In the thesis the Monte Carlo Tree Search, Minimax and Alpha-beta pruning algorithms are com- pared. The thesis also contains comparison of multiple evaluation functions. Few most common simplified versions of chess and also Fischer chess are available in the application. The theoretical part of the thesis contains basic information about chess and chess training, history of computer chess and description of used algorithms. There is also a comparison of implemented algorithms both according to their level of play and according to their usa- bility for training purposes. One of the interesting outcomes of the thesis is an observation that during the game the value of different figures changes and that in endgames the classical chess evaluation function is worse than evaluation function produced by an evolutionary algorithm. 1
|
|
|
Evolutionary Algorithm-based Procedural Level Generator for a Rogue-like Game
Vegricht, Jan ; Gemrot, Jakub (advisor) ; Zelinka, Mikuláš (referee)
Title: Evolutionary Algorithm-Based Procedural Level Generator for a Rogue-like Game Author: Jan Vegricht Department: Department of Software and Computer Science Education Supervisor: Mgr. Jakub Gemrot, Ph.D., Department of Software and Computer Science Education Abstract: Rogue-like games are genre with long tradition in game industry. One significant factor commonly associated with this genre is procedural level generation. The goal of this thesis is to design and implement a level generator for one concrete rogue-like game using evolutionary algorithms as main means of generation. Methods and results are then compared to non-evolutionary alternative algorithms, attempting to generate comparable solutions. The results seem to indicate that while evolutionary algorithms can be used to generate dungeons, practicality of this approach is for the most part limited. Keywords: evolutionary algorithms, procedural generation, constrained optimization, rogue-like
|
|
|
Using reinforcement learning to learn how to play text-based games
Zelinka, Mikuláš ; Kadlec, Rudolf (advisor) ; Lisý, Viliam (referee)
The ability to learn optimal control policies in systems where action space is defined by sentences in natural language would allow many interesting real-world applications such as automatic optimisation of dialogue systems. Text-based games with multiple endings and rewards are a promising platform for this task, since their feedback allows us to employ reinforcement learning techniques to jointly learn text representations and control policies. We present a general text game playing agent, testing its generalisation and transfer learning performance and showing its ability to play multiple games at once. We also present pyfiction, an open-source library for universal access to different text games that could, together with our agent that implements its interface, serve as a baseline for future research.
|
|
|
Using yaPOSH for CTF team behaviour
Zelinka, Mikuláš ; Bída, Michal (advisor) ; Bošanský, Branislav (referee)
Title: Using yaPOSH for CTF team behaviour Author: Mikuláš Zelinka Department: Department of Software and Computer Science Education Supervisor: Mgr. Michal Bída Abstract: We evaluated the suitability of yaPOSH (an action-selection system for virtual agents) for complex team behaviour development, specifically for the Capture the Flag (CTF) mode in Unreal Tournament 2004. We created a CTF team using yaPOSH and Java and compared them with a CTF team written by V. Tuma in plain Java as well as with the native UT2004 bots. We found out that although yaPOSH does have some advantages over plain Java (mainly in code readability), one cannot create a competitive bot using yaPOSH only. That is a direct consequence of the limitations the yaPOSH planner has, with the most significant one being lack of parallelism support. Thus, some as- pects of the behaviour (such as combat) were programmed in Java. Nevertheless, the resulting team is better than the native bots by a margin and slightly better than Tuma's CTF team. As a result, we have made several suggestions for improvements of the yaPOSH engine as well as its editor, such as to add the possibility to execute multiple ya- POSH plans at once, or to enable their parallel evaluation. These suggestions and our other findings are not limited to the chosen domain of...
|
guest ::
