Machine learning in trading: theory, models, practice and algo-trading - page 3085
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
I have tried to train EAs from this cycle using genetic and evolutionary methods (articles 30 and 31). The author has 1000 epochs in the parameters. The population is 50 individuals per epoch, as I understand. The best result is displayed in the log when training. So for 200 epochs this best result has not changed in relation to the initial one. I also put the population of 100 individuals and trained about 150 epochs. The effect is the same. So I gave up this method and moved on to newer ones.
And the newer ones are not trained properly? ) honestly, I don't want to read 30+ articles with no normal backtests.
what is this? from the last article :)
Are the newer ones not trained properly? ) honestly, no desire to read 30+ articles where there are no normal backtests
what is this? from the last article :)
In this article (and in the previous one) the author has implemented an algorithm of an ensemble of models (10 agents + 1 critic + 1 scheduler). It first collects a database of examples (first by random neurons, then the trained ones are used) and then another advisor trains the model on this database. According to the author's statement, the "scheduler" should use all agents evenly. But in my case it uses either one agent (and does nothing with it) or two (just opens one trade and waits until the test time runs out or balance).
I don't understand how to make it use all agents. In general, the author in the article (42) writes that he also had such a problem with the variant from (41) but in (42) he corrected it and now he uses all agents. .... Well, I can't force the model to use all agents. However, during all the experiments a couple of times the model accidentally used 3 or 4 agents and it was actually able to close trades and sort of trade.
PS: I asked the author under the article in the comment to explain and write in detail in figures how he trained the model. How many examples did he collect in the database? How many training passes did he make? At least some specific data on what to focus on..... But for some reason the author ignores the question. Although what is the point of writing a whole series of articles if it is impossible to use anything from it? If you think that self-promotion? But the author doesn't sell anything, then why does he need advertising....
In this article (and in the previous one) the author has implemented an algorithm of a kind of model ensemble (10 agents + 1 critic + 1 scheduler). It first collects a database of examples (first by random neurons, then the trained ones are used) and then another advisor trains the model on this database. According to the author's statement, the "scheduler" should use all agents evenly. But in fact it uses either one agent (and does nothing with them) or two (just opens one trade and waits until the test time runs out or balance).
I don't understand how to make it use all agents. In general, the author in article (42) writes that he also had such a problem with the variant from (41), but in (42) he fixed it and now he uses all agents. .... Well, I can't force the model to use all agents. However, during all the experiments, a couple of times the model accidentally used 3 or 4 agents and it was actually able to close trades and sort of trade.
PS: I asked the author under the article in the comment to explain and write in detail in figures how he trained the model. How many examples did he collect in the database? How many training passes did he make? At least some specific data on what to focus on..... But for some reason the author ignores the question. Although what is the point of writing a whole series of articles if it is impossible to use anything from it? If you think that self-promotion? But the author doesn't sell anything, then why does he need advertising....
Perhaps the author is so full of himself (since the work done is really colossal) that he no longer answers questions from mere mortals :).
unfortunately, I have no opportunity to check all this, as I am interested in other things.
I have only expressed my experience of using RL: the results on new data were mediocre, this is because these algorithms do not take into account all sorts of shifts and drifts and it is not quite clear how to incorporate such information there.
Maybe there are some solutions among the articles, I haven't looked into it.
Friends, hello!
There is a battle, welcome, make some noise!!!
...
Hello JeeCi! Good day to you!
I invite you to the championship, take part and have fun).