Discussing the article: "Neural Networks in Trading: Controlled Segmentation (Final Part)"
Dmitry hello. I got this error during training:
2024.10.08 21:28:01.820 Study (EURUSD,H1) RefMaskAct.nnw 2024.10.08 21:28:01.896 Study (EURUSD,H1) RefMaskCrt.nnw 2024.10.08 22:48:49.440 Study (EURUSD,H1) Train -> 294 -> Actor 0.0803357 2024.10.08 22:48:49.440 Study (EURUSD,H1) Train -> 295 -> Critic 0.0005726 2024.10.08 22:48:49.440 Study (EURUSD,H1) ExpertRemove() function called 2024.10.08 22:48:49.558 Study (EURUSD,H1) 14 undeleted dynamic objects found: 2024.10.08 22:48:49.558 Study (EURUSD,H1) 14 objects of class 'CBufferFloat' 2024.10.08 22:48:49.558 Study (EURUSD,H1) 19968 bytes of leaked memory found
What does it mean?
By the way, when compiling these 2 warnings appear:
Series.mqh ArrayDouble.mqh 'NeuroNet.cl' as 'const string cl_program' 1 deprecated behavior, hidden method calling will be disabled in a future MQL compiler version NeuroNet.mqh 30478 22 deprecated behavior, hidden method calling will be disabled in a future MQL compiler version NeuroNet.mqh 30700 22 code generated 1 0 errors, 2 warnings, 6344 msec elapsed, cpu='X64 Regular' 3
The files from the article are unchanged.
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 in Trading: Controlled Segmentation (Final Part).
Model training is performed offline. However, to maintain the training dataset relevance, we periodically update it by adding new episodes based on the current Actor policy. Model training and dataset updates are repeated until the desired performance is achieved.
During the preparation of this article, we developed a rather interesting Actor policy. The results of its testing on historical data from January 2024 are presented below.
The test period was not included in the training dataset. This testing approach simulates real-world model usage as closely as possible.
During the test period, the model executed 21 trades, 14 of which were profitable, which amounted to more than 66%. Notably, the proportion of profitable trades exceeded losing ones in both short and long positions. Moreover, the average profit per winning trade was twice the average loss per losing trade. The maximum profit trade was nearly three times greater than the largest loss. The balance chart shows a clearly defined upward trend.
Author: Dmitriy Gizlyk