Discussing the article: "Neural networks made easy (Part 43): Mastering skills without the reward function"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Neural networks made easy (Part 43): Mastering skills without the reward function.
The problem of reinforcement learning lies in the need to define a reward function. It can be complex or difficult to formalize. To address this problem, activity-based and environment-based approaches are being explored to learn skills without an explicit reward function.
To test the performance of the trained model, we used data from the first two weeks of May 2023, which was not included in the training set but closely follows the training period. This approach allows us to evaluate the performance of the model on new data, while the data remains comparable, since there is no time gap between the training and test sets.
For testing, we used the modified "DIAYN\Test.mq5" EA. The changes made affected only the data preparation algorithms in accordance with the model architecture and the process of preparing source data. The sequence of calling direct passes of models has also been changed. The process is built similarly to the previously described advisors for collecting a database of examples and training models. The detailed EA code is available in the attachment.
As a result of testing the trained model, a small profit was achieved, with the profit factor of 1.61 and the recovery factor of 3.21. Within the 240 bars of the test period, the model made 119 trades, and almost 55% of them were closed with a profit.
Author: Dmitriy Gizlyk