A quick and free library for MT4, much to the delight of neuralnetworkers

New comment
 

The code and description of the library can be found in the article: Using Neural Networks in MetaTrader


Thanks to the author!


The library proved to be working.


Here are the first results:



First results


The last 251 trades are history-fitting. All the rest on the left are OOS.


Of course the Expert Advisor had to be redesigned, i.e. I removed filtering to eliminate the hard fit. Redesigned it so that EA was in market all the time. Entries are wrong but even with them you can find a successful forward.


I attached the code of the Expert Advisor for those who are interested.


There is only one modifiable input parameter in the code: StopLoss. Set an optimization from 10 increments 1 to 100 for 4 digits and from 100 increments 10 to 1000 for 5 digits. Optimization is run without the genetic algorithm on a limited portion of history - just a few seconds. Then we switch off using the date and run it on the whole available history. We look for the most successful forwards.


The most important thing is that you can use this library very easily! There is nothing complicated about it.
Files:
nm.mq4  8 kb
 

The graph is not interesting. It looks like a random trade.

 
lea >> :

The graph is not interesting. It looks like a random trade.

I didn't expect to see anything similar to profitable from the first time on foul inputs even on the fit.


Neuralnetworkers know, that very often you have to wait long and painfully, sometimes hours, sometimes days, sometimes weeks, for the first results.


And sometimes, having waited for nothing and having spat on all this stuff, they write in this forum: that allegedly neural networks are rubbish, you just wasted your time.

 
Where is the corrected martin?
 
And fann is the theme in general. :) It's not a bad library for neuronics.
 
Red.Line >> :
And the fann is the theme in general. :) It's a very good library for neurons.

Of course it is! An adaptive committee of 16 meshes on 30 inputs rules in seconds.

 
Reshetov писал(а) >>

You can get the code and description of the library in the article: Using Neural Networks in MetaTrader

Thanks to the author!

The library seems to be working.

Here are the first results:


First results

The last 251 trades are history-fitting. Everything else on the left is OOS.

Of course EA had to be redesigned, i.e. I removed filtering to eliminate hard fitting. Redesigned it so that EA was in market all the time. Entries are wrong but even with them you can find a successful forward.

I am attaching the source code of the Expert Advisor for those who are interested.

There is only one modifiable input parameter in the code: StopLoss. Set an optimization setting from 10 in 1 to 100 increments for 4 characters, or from 100 in 10 to 1000 increments for 5 characters. Optimization is run without the genetic algorithm on a limited portion of history - just a few seconds. Then we switch off using the date and run it on the whole available history. Then we look for the most successful forwards.

The most important thing is that you can use this library very easily! There is nothing complicated about it.

Thanks for info and thanks to the author for creating the library. If somebody manage to create a working code using this funn

- Evolving topology training which dynamically builds and trains the ANN (Cascade2)

Please share.

 
gpwr >> :

Thanks for the info and to the author for creating the library. If someone manages to create a working code using this function

- Evolving topology training which dynamically builds and trains the ANN (Cascade2)

Please share.

And if somebody would throw up a link with sources of recurrence grid, it would be even more wonderful. ;-)

 
marketeer >> :

And if someone would throw up a link to the sources of the recurrence grid, that would be even better. ;-)

http://www.codeproject.com/ - for all tastes and colours

 

Here's a little more work on the inputs and the picture is better now. The last 198 trades are a fit, the rest on the left are OOS.



In short, it's clearly worth the effort.

 
Reshetov >> :



Here's a little more work on the inputs and the picture is better now. 198 last trades - fit, the rest on the left - OOS.



In short, it's clearly worth the effort.


Dear Sir! Please share the corrected source code. Have you tried to send prices of different currency pairs to the NS input? If so, what was the result?
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657
New comment
Reason: