1Likes
LinkBack URL
About LinkBacks
more...The subject of Kohonen neural networks was approached to in some articles on the mql5.com website, such as Using Self-Organizing Feature Maps (Kohonen Maps) in MetaTrader 5 and Self-Organizing Feature Maps (Kohonen Maps) - Revisiting the Subject. They introduced readers to the general principles of building neural networks of this type and visually analyzing the economic numbers of markets using such maps.
However, in practical terms, using Kohonen networks just for algorithmic trading has been confined with only one approach, namely the same visual analysis of topology maps built for the EA optimization results. In this case, one's value judgment, or rather one's vision and ability to draw reasonable conclusions from a picture turns out to be, perhaps, the crucial factor, sidelining the network properties regarding representing data in terms of nuts-and-bolts matters.
In other words, the features of neural network algorithms were not used to the full, i.e., they were used without automatically extracting knowledge or supporting decision making with specific recommendations. In this paper, we consider the problem of defining the optimal sets of robots' parameters in a more formalized manner. Moreover, we are going to apply Kohonen network to forecasting economic ranges. However, before proceeding to these applied problems, we should revise the existing source codes, get something fixed, and make some improvements.
It is highly recommended to read the above articles first, if you are not familiar with the terms such as 'network', 'layer', 'neuron' ('node'), 'link', 'weight', 'learning rate', 'learning range', and other notions related to Kohonen networks. Then we will have to saturate ourselves in this matter, so re-teaching the basic notions would lengthen this publication significantly.
Reply With Quote
Bookmarks