The Evolution Of NN with C#

July 22nd, 2009, 10:37 PM

I know some of you guys are using AForge, they released 2.0 final version today so you might want to check that out =)

aforge - Project Hosting on Google Code

August 2nd, 2009, 03:55 PM

Quantlib for C# == QLNet

QLNet | Get QLNet at SourceForge.net

August 3rd, 2009, 03:29 AM

does anyone have sample code on how you are using this in ninja?

August 3rd, 2009, 04:45 PM

hi sefstrat,

your input for building your system came from forex-tsd like Better NN EA development - Forex Trading

August 3rd, 2009, 09:51 PM

wh

hi sefstrat,

your input for building your system came from forex-tsd like Better NN EA development - Forex Trading

Not really, I have looked through that thread before a little bit.. a few good links but mostly just people throwing around ideas. I did find the cupsbook.pdf on that thread I believe which has some good examples of how to use a'trous wavelet technique (which is used on one of the inputs of my NN).

The NN part of my system is primarily based on the book I linked to earlier in this thread (NN control of nonlinear discrete time systems). It is a recurrent, closed-loop network.. meaning no offline training, all training is performed online in real time via convex optimization routines. Therefore it is inherently adaptive and not subject to many of the problems associated with supervised learning such as over-fitting, poor choice of sample data, etc..

Closed loop network is more difficult to implement and has its own set of challenges but in general they are explicit rather than implicit (implied from training data) so you can more easily control its mechanism of adaption and tune it the way you want.

September 7th, 2009, 08:41 AM

wh

For three weeks, I have communicated with neural networks and genetic coding.
Have heard about this for years, but ... nothing for me. Now the circle is closed.
As a seasoned Googler, I found this AForge.NET :: Computer Vision, Artificial Intelligence, Robotics Framework, which has beautiful examples of application. What important is for the learning process. See links below.

Last week i order a book by amazon: Introduction to Neural Networks for C#, 2nd Edition (Perfect Paperback)

Why i wrote this:

if(ninjatrader == c# && Afroge == c#)
bool isGood = true;

you can implement classes to your ninjascript or build external programms that you can connect. An other importent reason is: Aforge is open source and this is good. Free world, free sources and free money ...

Some good links for learning and understanding are:
CodeProject: Genetic Algorithm Library. Free source code and programming help
CodeProject: Evolution computations on C#. Free source code and programming help
CodeProject: AI : Neural Network for beginners (Part 1 of 3). Free source code and programming help
CodeProject: Neural Networks on C#. Free source code and programming help

The examples works well, and you can use your data for testing.
Create small indicator how write your close[i] or ma[i] or something else to a csv and then import csv to examples.

Arne

Hi wh,

I was thinking of toying with this again. I was wondering if you have a very basic NT framework for aforge you would share with me (just based on Close[0] is fine or whatever you have).

Mike

September 7th, 2009, 02:19 PM

hi mike here the samples from https://www.codeproject.com/. For more info go in and search on this site "Neural Network". There are a lot of good examples.

for data input i use a simple loop and copy & paste from output window in a file, but you can also generate auto a csv script. but i am fool. here a dirt sample.

Code

                            
if(CurrentBar < 48) return;
            
            double[] _close = new double[48];
            
            for(int i = 0, j = _close.Length; i < ma1.Length; i++)
            {
                _close[i] = Close[j];
                
                Print(_close[i]);
                j--;
                
            }
            Print("");

Also go here in

Seeing into the future...with Cycles.

True, wavelets are not that useful for discretionary trading, it is very powerful for algorithmic trading due to its power of feature extraction which makes pattern recognition much easier.

I skimmed through the research paper you linked above, only …

September 8th, 2009, 02:38 AM

wh

hi mike here the samples from https://www.codeproject.com/. For more info go in and search on this site "Neural Network". There are a lot of good examples.

for data input i use a simple loop and copy & paste from output window in a file, but you can also generate auto a csv script. but i am fool. here a dirt sample.

Code

                            
if(CurrentBar < 48) return;
            
            double[] _close = new double[48];
            
            for(int i = 0, j = _close.Length; i < ma1.Length; i++)
            {
                _close[i] = Close[j];
                
                Print(_close[i]);
                j--;
                
            }
            Print("");

Also go here in

Seeing into the future...with Cycles.

True, wavelets are not that useful for discretionary trading, it is very powerful for algorithmic trading due to its power of feature extraction which makes pattern recognition much easier.

I skimmed through the research paper you linked above, only …

Thanks Arne

Mike

September 8th, 2009, 07:24 AM

wh

hi mike here the samples from https://www.codeproject.com/. For more info go in and search on this site "Neural Network". There are a lot of good examples.

for data input i use a simple loop and copy & paste from output window in a file, but you can also generate auto a csv script. but i am fool. here a dirt sample.

Code

                            
if(CurrentBar < 48) return;
            
            double[] _close = new double[48];
            
            for(int i = 0, j = _close.Length; i < ma1.Length; i++)
            {
                _close[i] = Close[j];
                
                Print(_close[i]);
                j--;
                
            }
            Print("");

Also go here in

Seeing into the future...with Cycles.

True, wavelets are not that useful for discretionary trading, it is very powerful for algorithmic trading due to its power of feature extraction which makes pattern recognition much easier.

I skimmed through the research paper you linked above, only …

Ok, I was hoping someone had actually integrated it with Ninja already (no copy/paste). I was looking to feed it a series of variables and values and get back a prediction for a certain series.

Mike

September 8th, 2009, 11:44 AM

I am looking at SVM (Support Vector Machines). Specifically as implemented at
https://www.csie.ntu.edu.tw/~cjlin/libsvm/

I am trying out the C# version located at
Matthew Alastair Johnson - SVM.NET

SVM seems primarily for classification systems but it can also be used for regression. My initial impressions after getting a C# console version up and running is that real time training is not feasible. The training will have to be done offline in the background. There are sufficient methods in the C# implementation that real time predictions will be an easy implementation in Ninja from a model that is generated offline.

Instead of wavelets, look at SVD (Singular Value Decomposition). A C# implementation can be found at
Singular value decomposition - ALGLIB

I have run the test console version and I feel that this is something that can be done in real time. The only problem is that what is provided at the above link converts your input into an SVD representation. There is no code showing how to take this SVD internalization to produce output. This should not be a big problem but I do not have the math skills to determine the code to reverse the SVD back into the real world. Maybe there is someone out there who can do this. Should be similar to PCA.

The Evolution Of NN with C#

Discussion in NinjaTrader

The Evolution Of NN with C#