Machine learning for robots - page 2

 
Ivan Negreshniy:
As for signals, there the template can be obtained directly from the terminal panel, by <Show on chart> button, and by available Expert Advisor, after testing with visualization enabled and both these special templates are quite suitable for code generation, but as I mentioned above, this is a slightly different topic.

I can't find similar mentions/explanations, can you repeat that? Why is this not suitable?

Ivan Negreshniy:

Because it is preferable not to generate new EAs on the basis of such templates, but filters to improve the original EAs, so I don't want to bring this up yet, as it would be unnecessary confusion.

I don't understand the difference between hand-drawn arrows or by the result of a pro-trade.

I am not picking on you, I just want to help you develop the subject, as there is a misunderstanding, and I think I am not the only one.

 
Aleksey Vyazmikin:

I can't find similar mentions/explanations, can you repeat that? Why doesn't it fit?

I don't see what the difference is, hand-drawn arrows or by the result of pro-trading.

Not picking on you, just want to help you develop the topic, as there is a misunderstanding, and I think I'm not the only one.

You can see distinctive features of special patterns from different points of view and I will try to show them to you, let's start with the legal one.

For instance, you have copied the chart template of a good trading session or a visual report of the trial version of an expensive EA in the Market service with one click and I have created a surprisingly good clone of it. What have we got as a result - scandals and complaints from the EA's authors and a ban from the EA's administration.

Of course, it is possible that you yourself are the author of that EA, whose trade you made a template and it is easy to prove, because it is the fruit of your many years of work.
But from a technical point of view, it means that trading of your EA is probably a whole set of different techniques and methods, such as trawls, averaging, grids, locks, etc., which in principle can also be modeled, but for this they need to be formalized in a strategy pattern and model of MP, otherwise, most likely, we will get a banal sinker.

Finally, let us also assume that you promptly formalized your multiple and original trading elements, and I specifically for you, modeled them into additional neural components and generated a unique Expert Advisor based on them.
However, the result is again questionable from the scientific point of view, it is not a pure experiment where a lot of unproven elements are tested simultaneously and no usefulness is likely to come from it.

And in general, how can we discuss such experiments if everyone has his own, original set of components, IMHO it would be like a well-known MO forum thread where everyone has something and it seems to work somehow, somewhere, but there's nothing to show, discuss, let alone to exchange with someone ...))

 
Aleksey Vyazmikin:

In short, my IMHO opinion, with the above in mind - there is no point in cloning ready-made EAs with MO, instead they can and should be improved, e.g. with neurofilters.

And I, in an attempt to liven up my experiments, added a signal indicator filter to makeSignals, a MACD template creation utility.

I started with a primitive strategy - sell when the signal line is higher than the main one, buy when it is lower, default setting is MACD(12,26,9).

I used this indicator as a filter to create a template, for this I ran makeSignals (see attachment), set the settings (see the template) and no further action but save(Charts/Schedule/Save).

The template contains more than two thousand training signals from the beginning of 2011 till the end of May 2018, the summer period from June till the current moment is left for the forward test.

Using the specified template, I have developed two Expert Advisor variants, MACD_EURUSD_H1_NN - based on the neural network and MACD_EURUSD_H1_RF - on random forests (see the attachment).

Expert Advisors are tested on EURUSD H1 server MetaQuotes-Demo, on the test chart of one of the Expert Advisor is marked training interval boundary.

Thus, the test shows that outside of the training, the EA continues to trade in the plus side in addition, the following screenshot shows that the MACD signal logic continues to be followed in the unmarked section of the template as well.

Files:
 

Any closed code is considered in two aspects: an attempt to advertise a paid code and hidden code capable of causing harm on the computer being run.

Advertising, much less causing harm, is forbidden. It should also be remembered that the forum serves an educational purpose - and for this the code should only be open source.

 
Vladimir Karputov:

Any closed code is considered in two aspects: an attempt to advertise a paid code and hidden code capable of causing harm on the computer being run.

Advertising, much less causing harm, is forbidden. Also, don't forget that the forum serves an educational purpose - and for that code should only be open source.

Well, this topic says that the code is made free of charge using the author's template and only the author will receive the source code and be able to publish it. The rest of us can test and discuss compiled EAs on MetaQuotes-Demo and this, given the sandbox protection, excludes any damage.

You see that even with these conditions people are afraid or have not yet figured out how to draw their strategies, I have only made examples and a description, and you are changing the rules and cutting the topic on its roots, what kind of educational goals are there, rather the opposite.

 
Ivan Negreshniy:

So, the topic says that the code is made for free, by the author's template and only the author will receive the source code and be able to publish it, the rest can test and discuss compiled EAs on MetaQuotes-Demo, and this, given the sandbox protection, excludes any damage.

You see for yourself that people even with such conditions are wary or have not yet figured out how to draw their strategies, I have only made examples and description, and you change the rules and chop the topic at the root, what educational goals there, rather the opposite.

Don't even try to promote advertising under plausible pretexts. The discussion is over.

 
Vladimir Karputov:

Don't even try to promote advertising under plausible pretexts. The discussion is over.

In general, I think the project is a dead end, but to ban it is nonsense. The man proposes, he wants to do it - let him do it. Why bother?

Right now he is doing it for free. And whether he will pay - this is still a fork in the water. If it does, then ban it.

 

Hi!

Trained my bot the other day... and what do you think?

The bot said: "I won't trade forex, it's impossible to predict anything! trade with your hands yourself!"

All the words are intuitively typed by himself through the AI.


p.s.yyyy

 
Yuriy Asaulenko:

In general, I think the project is a dead end, but to ban it is nonsense. The person who proposes it wants to do it, let him do it. What are you pestering me for?

Right now he is doing it for free. Will he be paid for it, it's still a hair on the bucket. If it does, then ban it.

Thanks for the support, and as for the deadlock of the project, the experiment just serves to determine the prospects, especially for a brand new, freshly patented technology, but as for hackneyed, Python and ERA libraries, that's another matter.
 
I've read the forum thread 2 times and still don't understand what you need. Please send me the source code, send me a description of the model, give me some room for experiments. I was training a neural network using MACD signals but I didn't see an impressive model. It would be interesting to see the model from your side.
Reason: