Discussion of article "Neural networks made easy (Part 35): Intrinsic Curiosity Module"
The balance line is strange. It seems that during the year only deals were opened and at the end of testing they were all closed. All trades on sale and Euro all year in the fall. Have been sitting out profits and losses for almost a year).
The balance line is strange. It seems that during the year only deals were opened and at the end of testing they were all closed. All trades on sale and Euro all year in the fall. Have been sitting out profits and losses for almost a year).
Agreed. But this is just a demonstration of the technology. The model requires longer training to work properly. The screenshots posted are only from the 2nd epoch of training. Which is very little.
Hi Thank you for your awesome article.
I am eager to do some testing, but I have difficulty in creating the models even you mentioned " To train the Expert Advisor, all models were created using the NetCreator tool . "
What should I do? thanks.
- www.mql5.com
Please help
Thanks
Very successful article. Could you possibly help me to create the models? neuron numbers, which layers, with which activation, with which outputs. Thank you in advance for your help.
Do you have the model file? it seems not in the zip file.
Do you have more information about how to create the model by the NetCreator as well or at least share this file? the EA can't start run withtout those file.
as said below:
To train the EA, all models were created using the NetCreator tool. It should be added that to enable EA operation in the strategy tester, the model files must be located in the terminal common directory 'Terminal\Common\Files', since each agent operates in its own sandbox, so they can exchange data only via the common terminals folder.
- www.mql5.com
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
New article Neural networks made easy (Part 35): Intrinsic Curiosity Module has been published:
We continue to study reinforcement learning algorithms. All the algorithms we have considered so far required the creation of a reward policy to enable the agent to evaluate each of its actions at each transition from one system state to another. However, this approach is rather artificial. In practice, there is some time lag between an action and a reward. In this article, we will get acquainted with a model training algorithm which can work with various time delays from the action to the reward.
To train the EA, all models were created using the NetCreator tool. It should be added that to enable EA operation in the strategy tester, the model files must be located in the terminal common directory 'Terminal\Common\Files', since each agent operates in its own sandbox, so they can exchange data only via the common terminals folder.
Training in the strategy tester takes a little longer than the previous virtual training approach. For this reason, I reduced the model training period to 10 months. The rest of the test parameters remained unchanged. Again, I used EURUSD on the H1 timeframe. Indicators were used with default parameters.
To be honest, I expected that the learning process would begin with the deposit loss. But during the first pass, the model showed a result close to 0. Then it even received some profit in the second pass. The model performed 330 trades with more than 98% of operations being profitable.
Author: Dmitriy Gizlyk