Artificial neural networks. - page 13
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
yes, the class_NetMLP.mqh library in the folder ....\MQL5\Include
in the meta editor create a script and copy the code
hmm... Well the result is already known - NS will learn multiplication table, try to teach, for example, sin(x) - in a word experiment to come to an understanding why, and how it works, in principle, does not matter - but it sure works. I think the main thing for you is to learn how to use a tool like NS.
:-) I need to learn how to turn this thing on, I drop a script on the chart, it loads and unloads....
What or how to press to see what's going on
I want to understand how I can apply it.
The question is purely theoretical, is it possible, using the network, on the example of a particular trader's work, to try to teach the Expert Advisor to make trades, as a learning by feeding trades to the network, plus and most "correct", and excluding erroneous deals, loss-making and ambiguous ones?
---------------
And explain how to enable what I compiled. please.
look here, there's also the output via "print":
Theoretical question, is it possible, using the network, on the example of a particular trader's work, to try to teach an Expert Advisor to make transactions, as a learning by feeding the network deals with plus and most "correct", and excluding erroneous deals, losing and ambiguous?
this is not a theoretical question, but a really practical one... NSs can remember the structure of the input data when trained and then the trained NS will produce the correct output responses
but it's not all that smooth... the main problem is what to give to the input of the NS, there are trivial mistakes like: we teach NS the multiplication table 1x1 ... 9x9, and then we ask NS correct answer 23x13 and complain that NS do not work - NS just not trained in multiplication table 23x13.
If we have decided that using 3-4 last Close[] for EURUSD we can predict where the price will go in 10 bars and for a long time "torment" the NS in this direction and then shouting at the forums, NS doesn't work... (although you should probably use moon phases for prediction :) )
I.e. quality of NS performance depends on properly prepared data, if there are hidden dependencies, NS learns them and will work correctly in the future, if there are no dependencies, NS cannot do miracles
That's the way it is, I can't do it scientifically otherwise.
see here, there's also the output via "print":
А... and I've got this
here's what I have left on this library - do not even know if it will help you or not
SZS:I updated my previous post, now gone, business - finally, my scripts on the statistics gave the results, 14 hours of running the computer, I will learn
see here, there's also the output via "print":
this is not a theoretical question, but a really practical one... NSs can remember the structure of the input data when trained and then the trained NS will produce the correct output responses
but it's not all that smooth... the main problem is what to give to the input of the NS, there are trivial mistakes like: we teach NS the multiplication table 1x1 ... 9x9, and then we ask NS correct answer 23x13 and complain that NS do not work - NS just not trained in multiplication table 23x13.
If we have decided that using 3-4 last Close[] for EURUSD we can predict where the price will go in 10 bars and for a long time "torment" the NS in this direction and then shouting at the forums, NS doesn't work... (although you should probably use moon phases for prediction :) )
I.e. quality of NS performance depends on properly prepared data, if there are hidden dependencies, NS learns them and will work correctly in the future, if there are no dependencies, NS cannot do miracles
That's the way it is, I can't do it scientifically otherwise.
The problem of adequacy of training is that the input data must be correct if we want to get the desired result.
And sufficiently formalized, and if indeed only on their basis a decision is made
My head is spinning around the idea that the input stream of data on the basis of which a decision is taken is not wide, something like twenty, thirty million combinations +- one order of magnitude, according to my rough guess, after training will be two or three thousand combinations
I'm trying to dig in that direction.
here's what I have left on this library - do not even know if it will help you or not
SZS:I updated my previous post, now I'm off, business - finally my scripts on the statistics gave the results, 14 hours of running the computer, I will study
There are cases when the input data doesn't depend on the output data, i.e. we decided that using 3-4 last Close[] for EURUSD we can predict where the price will go in 10 bars and we "mess" with NS in this direction and then shouting at the forums that NS doesn't work... (although you should probably use moon phases for prediction :) )
I.e. quality of NS performance depends on properly prepared data, if there are hidden dependencies, NS learns them and will work correctly in the future, if there are no dependencies, NS cannot do miracles
That's the way it is, I can't do it scientifically.
Yeah, like that...
That's the main problem when training a neural network. As, indeed, in life in general. We never know what we might need in the future, we try to learn everything and anything we can get our hands on, and fate, the bitch of it all, says "What the hell did you learn all that for? You should have learned it... Anyway, whoever guesses, wins. Just like the master genetic algorithm says.