"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 24
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
Too bad.)) This is the program at the moment the maximum for users of all levels.
Uh-huh, so inviting Leov would be cool. Or someone else, who has already worked on NSDT.
I read this thread and now at least I'm beginning to imagine how complicated everything is.)
I can post a practically finished code for 4k using SOM.
At least to make it clear what I mean.
On the proposed one. The inputs and outputs are separated from the network. By any means. Their formation can be automated. But this is also not an easy task and only for typical variants.
A little aside, you'll need some coding, to one degree or another. But most often at a very trivial level.
The proposed scheme of learning -- OOS is inefficient. I've said it before, and more than once I think.
The market is not an area where you can check the quality of the network on the feedback. The only way to check the network quality is to test it on the training sample. And overtraining is removed simply - we reduce the number of tuning parameters to the minimum teachable.
TheXpert:
And overtraining is removed simply by reducing the number of settings to the minimum teachable.
I can post a practically finished code for 4k using SOM.
At least to make it clear what I mean.
That would be good, maybe you already have it handy.
As suggested. The inputs and outputs are separated from the network. By all means. Their formation can be automated. But it is also not an easy task and it is only for the typical variants.
A little bit aside, you'll need coding, to one degree or another. But most often at a very trivial level.
The generic options meet the needs of most users. That's what I suggest, so that there is a possibility to insert coding - your own preprocessing and postprocessing of data.
The proposed scheme of training -- OOS is inefficient. I've already said it more than once.
Market is not the area where you can check the quality of the network on the feedback. The only way to check the network quality is to test it on the training sample. You can remove the learning curve by reducing the number of tuning parameters to the minimum learnable.
And when you're trading in the real world it's also OOS, not a training sample. The OOS has disadvantages especially if used incorrectly, but it can be used. But it's probably better to ask potential users of these libraries - is it necessary or not. Or better just initially in the concept to make it possible to implement such methods.
It is strange to see here the posts of the person who claimed that neural networks and optimization algorithms are not applicable to the market....
or by increasing the number of Sample examples until the network is no longer effectively trained.
:) So you think that the more examples you give, the better? This is just not for trading. It is not that simple: you want to decrease the number of input parameters - you should feed exactly as much as you need, not to prevent re-learning. It's the same with the training period and the number of examples - since 1917 I want there to be more examples)))
As in the joke: "Drunkard lost his wallet and is looking for it under the lantern. They ask him:
- Where did you lose it?
- Somewhere over there.
- Why are you looking here?
- Because it's brighter here" :)
:) so you think that the more examples you give, the better? That's just not for tneiding.
It would be good, maybe you already have everything conveniently done.
There are three modes -- training, running, and auto-learning. For online trading now, it is only in the run mode.
The autotraining is just the OOS mode, on which you can check the prospects of the strategy.
Let me express my skepticism about the project. First of all, at the moment you cannot use even the library of basic trading functions and the wizard in real life without modifications. This is, so to speak, level 0 of twilight for the average non-programmer trader. Trading is now impossible for a non-programmer using the Expert Advisor of the wizard and the trading functions of the library. So what are we talking about? Building a spaceship without even building a wheel, a bicycle or other bricks? Sorry, if this is for the community of non-programmer traders, and not for a narrow group of enthusiasts - it does not look serious.
P.S.
If you need a code example for the method of differential evolution - I can send it to you. The method is pretty simple. It looks for global extrema.
P.S.
Can someone explain why we need NS and what they do. I've read Wiki, I don't understand much.