Discussion of article "Neural networks made easy (Part 30): Genetic algorithms"
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
New article Neural networks made easy (Part 30): Genetic algorithms has been published:
Today I want to introduce you to a slightly different learning method. We can say that it is borrowed from Darwin's theory of evolution. It is probably less controllable than the previously discussed methods but it allows training non-differentiable models.
he optimization process was tested with all previously used parameters. The training sample is the EURUSD H1 historical data. For the optimization process, I used the history for the last 2 years. The EA was used with default parameters. As a model for testing, I used architectures from the previous article with the search for the optimal probability distribution of decision making. This approach enables the substitution of the optimized model into the "REINFORCE-test.mq5" Expert Advisor used earlier. As you can see, this is the third approach in the process of training a model of the same architecture. Previously, we have already trained a similar model using the Policy Gradient and Actor-Critic algorithms. So, it is even more interesting to observe the optimization results.
When optimizing the model, we did not use the last-month data. Thus, we left some data for testing the optimized model. The optimized model was run in the Strategy Tester. It generated the following result.
As you can see from the presented graph, we got a growing balance graph. But its profitability is a bit lower than that obtained when training a similar model using the Actor-Critic method. It also executed less trading operations. Actually, the number of trades decreased by two times.
Author: Dmitriy Gizlyk