Archive of Scientific Articles

Abstract. The paper examines the problem of forecasting with the use of neural nets on the basis of the game "Ping-pong". Various configurations of neural nets and heuristics influencing learning speed and the error of the net are studied. For the visualization of the process of learning a special utility is created allowing the researcher to build various configurations of neural nets and watch the change of the net at separate stages of learning. The project is written in the Python language with the use of the library PyBrain.

Key words and phrases: нейрон, нейронные сети, метод обратного распространения ошибки, многослойный персептрон, язык программирования Python, Ping-pong, neuron, neural nets, method of reversal spread of error, multilayer perceptron, programming language Python, Ping-pong