Better NN EA development - page 7

 
tiger_wong:
Hello guys,

Please suggest me what the best algorithm that I should learn so I won't be wasting my time. From what I read here, there are several method using PNN, SVM, GPF, etc. Which one is the best? Maybe experienced person could help me.

Thank you

PNN is the simplest type of network (at least for implementation). It creates a node for each exemplar and then it compares the distances - nothing special happens at the nodes - just a distance measure passed through a rbf. Typically it's used for a small number of observations as the memory usage increases quickly. For continual re-training, as Better has it, it is probably best suited because, above everything else, the weights can be computed directly (analytically).

Hmm, I wonder whether Better wasn't just lucky to have a trending market. The same upward pointing patterns repeated and his system simply captured them near the optimal points...

I dunno nothing bout GPFs.

I suppose SVMs can be very powerful if you know what you are doing.

I did not.

Feed forward nets have given me good results.

Repeating what I said in a previous posting - it's all in the problem definition, not the algorithm.

-------------------------------

Don't take anything for granted.

I am not an expert.

 

Our group also offers a bridge from MT4 to Currenex through FIX API. We have a broker who agrees to give us Currenex pricing on Demo MT4 platform and we send orders from MT4 to FIX API to CNX in <30ms.

We are going to start working with neural nets and data mining for optimal moving average distances for entry and exits as a filter for our other systems.

 

MagicNumber = 855

Files:
nn15.jpg  85 kb
 

MagicNumber = 856

Files:
nn16.jpg  64 kb
 

Better EA orders with MagicNumber = 854

Files:
nn14.jpg  89 kb
betterea.zip  95 kb
 

Having not looked at the previously enclosed "BetterEA.zip," It seems we won't even want to mimic the system using the Magic Number, 856. System 854 definitely looks like profitable trades. I'll need to look a bit closer at System 855 to see how profitable it is, or if it's pretty close to break even. I'm actually super busy until next week, but I have some good ideas on engineering something that can accurately mimic what I've been referring to as System 854.

As a side note about doing Neural Networks in MT4, all of my personal, foreseeable applications of Neural Nets won't need a retraining phase any more often than once a week. Since one needs to do some MQL4 trickery for true continuous retraining AND that could potentially eat up too many resources on a given system, I personally plan on doing retraining phases no more than the once a week when my broker feed stops (possibly not even that often).

-MRE-

 

i've seen no full source code of ea here? mr barnix, please give me one?

thank u

look like i dont understand much what u'll disscusing all about but its interesting since 'better' won his champs in 2007...thumb ups

 

Orders distributed in time

Files:
nn17.jpg  82 kb
 

BUY & SELL Orders distributed in time 1

Signal1={buy,sell,buy,sell,buy,sell,buy,sell,buy,sell,buy,sell, ...}

Signal2={sell,buy,sell,buy,sell,buy,sell,buy,sell,buy,sell,buy ...}

Signal3={buy,sell,buy,sell,buy,sell,buy,sell,buy,sell,buy,sell, ...}

Files:
nn18.jpg  53 kb
 

5162668 2007.11.27 09:34 buy

Files:
Reason: