Implementace agenta Dyna-Q do testu algoritmického IQ

The Algorithmic IQ test is one of the attempts to create a method for evaluating the intelligence of artificial systems. It is based on the framework of reinforcement learning and algorithmic information theory, and it allows for an approximation of universal intelligence across many types of artificial agents. The aim of this thesis was to expand the implementation of the test by adding a fully functional Dyna-Q agent, which is a reinforcement learning agent that combines the Q-learning algorithm with planning using a model of the environment. The implementation was designed based on existing agents used in the test and the description of Dyna-Q in the literature. Python was used to implement and capture all the necessary functions of the agent. An initial evaluation of the agent was subsequently carried out using the AIQ test. The evaluation revealed several behavioural characteristics of this algorithm, such as better performance of the agent when using a lesser number of planning steps and a preference for higher discount factors with a lower probability of random action selection. However, due to the limited scope of the evaluation within the AIQ test, these findings may be incorrect or incomplete.