Discussion of article "Creating Neural Network EAs Using MQL5 Wizard and Hlaiman EA Generator" - page 3
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
Thank you all for your participation in the discussion and feedback, who is interested to see in detail the signals of the test EA Hlaiman EA Generator 007 -
Login : 1512007
Password : a3mlnkj
Server : MetaQuotes-Demo
A hammer is also a tool, but you can't split an atom with it. It's not that simple.
sometimes it's enough to read the comments to an article to.... not to waste time on the article itself.
thanks to the commenters. :) and, all the same, thanks to the afftar for his labour, thanks to MQ for the money spent on the afftar's labour. in short, "peace, labour, may!".
It's easy to build a bunch of neural networks, and it's easy to show how profitable they are on a backtest. But a forward test is needed to determine how these networks will behave on untrained data. The dynamics of market changes and the frequency of retraining networks have nothing to do with it. Retrain the network at every new tick, it won't help profitability on the real market. How to create such a network, which would sustainably bring profit on new data, is the most important thing. And the signals on the pipswitcher do not confirm anything. There are plenty of profitable demo pipers without neural networks.
...
It was sort of said (but I will repeat myself), for qualitative training of NS it is necessary to give it qualitative examples (not contradictory, with guaranteed presence of images), but only if you have an algorithm of preparation of such data you don't need NS (they can be described by other means).
The circle is closed.
I liked the article and the potential of the product. At least the man went from words to action.
Respect to the author!
It was kind of said (but I'll repeat it), for qualitative training of NS you need to give it qualitative examples (not contradictory, with guaranteed presence of images), but only if you have an algorithm of preparation of such data you don't need NS (they can be described by other means).
The circle is closed.
But you don't have to give up NS at all, because the data preparation algorithm you mentioned can be built on neural network components, self-organising Kohonen maps (SOM) or genetic algorithm (GA), for example in Hlaiman application there is even a separate plugin with such components and non-linear filters.
.
However, this article is not intended to prove or disprove the effectiveness of NSs, but merely proposes a means of automation for their simple application, based on MQL5 Wizard, as simple as classical indicators of technical analysis or in any combination with these indicators.
.
I liked the article and the potential of the product. At least the man went from words to action.
Respect to the author!
thanks for your participation in the discussion and feedback, who is interested to see in detail the signals of the test EA Hlaiman EA Generator 007 -
Login : 1512007
Password : a3mlnkj
Server : MetaQuotes-Demo
The above signal is a real-time forward test on a series of >4000 trades. The test Expert Advisor 007 uses an additional module for MQL5 Wizard SignalHFT.mqh, which is currently being tested and improved.
This module for MT5 can be provided to licensed clients only in the technical support and update mode. Experiments on high-frequency trading using Hlaiman EA Generator on MT4 have been discussed on this forum before.
The above signal is a real-time forward test on a series of >4000 trades. The test Expert Advisor 007 uses an additional module for MQL5 Wizard SignalHFT.mqh, which is currently being tested and refined.
This module for MT5 can be provided to licensed clients only in the technical support and update mode. Experiments on high-frequency trading using Hlaiman EA Generator on MT4 have been discussed on this forum before.
Interesting work. Interesting solutions.
Neither in the article nor in the description of the package I did not find: what input data does the network use during training?
Maybe I missed it somewhere.
The description of the package is very vague. Here you have "fuzzy logic" and "neural network" - it sounds beautiful. Can you be more specific?
To evaluate a product, you need to know
input data (indicators, statistics, etc.)
the number of them and how they are selected.
and of course, what networks are used, training methods and other details without which this "black box" will remain dark. And to use it in trading is an extreme.
Otherwise, the approach is interesting: "You don't need to know anything about THIS. Just switch it on and work".
Good luck.
Interesting work. Interesting solutions.
Neither in the article nor in the package description I didn't find: what input data does the network use during training?
Maybe I missed it somewhere.
The description of the package is very vague. Here you have "fuzzy logic" and "neural network" - it sounds beautiful. Can you be more specific?
To evaluate a product, you need to know
input data (indicators, statistics, etc.)
the number of them and how they are selected.
and of course, what networks are used, training methods and other details without which this "black box" will remain dark. And to use it in trading is an extreme.
Otherwise, the approach is interesting: "You don't need to know anything about THIS. Just switch it on and work".
Good luck.
It's a bygone stage with this gentleman(s). In the rapid-fire branch you can read, he is there - lohhft, asked about network architecture and data preprocessing and other only interesting details in this context (PBX developers). The comrade was stupidly frosty, moving away then trying to change the subject, then claiming that he is not a specialist in AI, then that in general it is not his product, and then in another topic saying that it is already his, etc.
If you go to his site and read his articles selectively diagonally, it becomes clear that to some extent he is right, that he is not an AI expert is a fact. Just a fictionalist with schizoid tendencies.
As for the article, the fact that it was published here proves once again that you know what....
But you don't have to give up NS at all, because the data preparation algorithm you mentioned can be built on neural network components, self-organising Kohonen maps (SOM) or genetic algorithm (GA), for example, Hlaiman application even has a separate plug-in with such components and non-linear filters.
.
However, this article is not intended to prove or disprove the effectiveness of NSs, but merely proposes a means of automation for their simple application, based on MQL5 Wizard, as simple as classic indicators of technical analysis or in any combination with these indicators.
.
On the highlighted, don't lead the reader into delusion, GA is not an implementation of NS, GA is a method of optimisation.
As for the fact that NS can be a tool for selecting examples, I don't disagree, but I don't confirm it either.
If we copy nature, it seems that it is NS that selects examples, but let's describe for a second the whole prehistory of man's emergence:
With the help of GAs, cells were selected that gave a competitive advantage, then let's skip the fact that it is easier to survive together and the emergence of organisms and go straight to the emergence of NS cells.
By then, cells are already born with built-in reflexes. That is, millions of years of GAs have picked up such an apparatus that successfully copes with the current tasks of survival. Then again we rewind and we have individuals who take care of their children and pass on information on the principle of "do as I do".
The principle is excellent, but the very information that the individuals possess is bred with the help of the same GA, those are very many years of evolution.
That's the end of the opus. Conclusion: the examples for the transmission of the principle "do as I do" were bred on a large population for a very long time. And this evolution continues, the world changes, leaders change and examples change.
Now let's get to our sheep: the forex market is relatively young, one, the human apparatus is not adapted for the processing of abstractions and numbers (as a realisation of abstraction), two, the human has a doheralion of nerve cells with a fucking billion connections and he can barely cope with the forex (it is not a fact that the current winners are not just random monkeys), three. And you want to repeat the million-year evolution on NS (well, say 100x100 neurons), with (well, say 1000 connections) ??? four.