Better NN EA development - page 22

 
 

thanks Barnix for all of your ressources

i have joined other document, simple one to understand the theory,

sorry it's for French people, but it is comprehensible for others to undertand pnn algorithm (rbf, noyau)

i work a lot on these algorithm, and it's working very well. I work on M5 with 100000 examples, and 10 inputs (ema M5, ema H1, ema D1, sto)

i've got very good results in real, near Better one, i use PNN and not SVM machine, difficult to implement SVM algo

but i want to improve the ea,

i work now on optimization of sigma, wich is a very important parameter, i optimize it by a lot of simulation

now i want to use rbf-dda algorithm to optimize it, thanks a lot for the rbf-dda.mq4 file found here , i'm going to integrate it in my EA

i will give you my results soon

if someone have already works on rbf-dda, tell me, the more we are, the best we can do

Files:
 
dmeliki:
Hello Barnix,

I have been working on PNN, found some implementations of it from fxreal.ru, have modified something, it works fine however, have to optimize sigma. I have been thinking if rbf-dda has been implemented, as far as understand it will find optimal sigma on its own during training process.

Here are the results of ea that works on pnn and trades on day close.

Strategy Tester Report

Atlantis[pnn]v2

InterbankFX-Demo Accounts (Build 216)

Symbol GBPUSDm (Great Britain Pound vs US Dollar)

Period Daily (D1) 2008.03.16 00:00 - 2008.05.15 00:00 (2008.03.16 - 2008.05.16)

Did you test it with the same data which you used for training? If not then how much training data did you use?

What kind of inputs and outputs did you use?

 
stef:

i work a lot on these algorithm, and it's working very well. I work on M5 with 100000 examples, and 10 inputs (ema M5, ema H1, ema D1, sto)

i've got very good results in real, near Better one, i use PNN and not SVM machine, difficult to implement SVM algo

Very interesting. I have been experimenting with NN and SVM. SVM performance is way less than NN but maybe I just have some problem in it. It seems that Barnix has made some progress with it but it is really hard to estimate results just from couple of screen shots. My SVM preforms really well on the training data but with test data it is just crap. NN is not too good either so I have not bothered to write EA yet.

Does your PNN perform best on M5 timeframe? Can you share any light on your outputs?

 

hi barnix any updates?

 

Strategy Tester Report for SVM_v_05_EURUSD EA

Settings:

//////////////////////////////////////////

#define NR_BARS1 2000

#define NR_BARS2 600

extern string Symb0 ="EURUSD";

extern double Lots =0.1;

extern int diff1 =true;

extern int decimalplace =14;

extern int svmtype =1; //0-libsvm,1-svmlight

extern int BasePeriod =PERIOD_M5;

extern int TrainPeriod =PERIOD_H4;

extern int class_type =2; //0-multiclass,1-class(0,1,-1), 2-class(-3,-2,-1,0,1,2,3)

extern int StopLoss =40; //30-50pips

extern int TakeProfit =40;

extern int FutureWindow =100;

extern int Signal =8;

extern bool UseRamDisk =true;

extern datetime BeginDate = D'2008.04.01 08:15';

//////////////////////////////////////////////////

///////////////////////////////////////

calc(r1,rv1);

calc(r2,rv1);

calc(r3,rv1);

calc(r4,rv1);

calc(r5,rv1);

calc(r6,rv1);

calc(r7,rv1);

calc(r8,rv1);

calc(r9,rv1);

/*

calc(r10,rv1);

calc(r11,rv1);

calc(r12,rv1);

calc(r13,rv1);

calc(r14,rv1);

calc(r15,rv1);

calc(r16,rv1);

calc(r17,rv1);

calc(r18,rv1);

calc(r19,rv1);

calc(r20,rv1);

calc(r21,rv1);

calc(r22,rv1);

calc(r23,rv1);

calc(r24,rv1);

calc(r25,rv1);

calc(r26,rv1);

calc(r27,rv1);

calc(r28,rv1);

calc(r29,rv1);

calc(r30,rv1);

*/

////////////////////////////////

 

Strategy Tester Report for SVM_v_05_EURUSD EA

SVM_train_M5 Settings:

 

Strategy Tester Report for SVM_v_05_EURUSD EA

Settings:

///////////////////////////////////////

#define NR_BARS1 2000

#define NR_BARS2 600

extern string Symb0 ="EURUSD";

extern double Lots =0.1;

extern int diff1 =true;

extern int decimalplace =14;

extern int svmtype =1; //0-libsvm,1-svmlight

extern int BasePeriod =PERIOD_M5;

extern int TrainPeriod =PERIOD_H4;

extern int class_type =2; //0-multiclass,1-class(0,1,-1), 2-class(-3,-2,-1,0,1,2,3)

extern int StopLoss =40; //30-50pips

extern int TakeProfit =40;

extern int FutureWindow =100;

extern int Signal =8;

extern bool UseRamDisk =true;

extern datetime BeginDate = D'2008.04.01 08:15';

/////////////////////////////////////////////////

///////////////////////////////////////

calc(r1,rv1);

calc(r2,rv1);

calc(r3,rv1);

calc(r4,rv1);

calc(r5,rv1);

calc(r6,rv1);

calc(r7,rv1);

calc(r8,rv1);

calc(r9,rv1);

/*

calc(r10,rv1);

calc(r11,rv1);

calc(r12,rv1);

calc(r13,rv1);

calc(r14,rv1);

calc(r15,rv1);

calc(r16,rv1);

calc(r17,rv1);

calc(r18,rv1);

calc(r19,rv1);

calc(r20,rv1);

calc(r21,rv1);

calc(r22,rv1);

calc(r23,rv1);

calc(r24,rv1);

calc(r25,rv1);

calc(r26,rv1);

calc(r27,rv1);

calc(r28,rv1);

calc(r29,rv1);

calc(r30,rv1);

*/

////////////////////////////////

Files:
 
viktor2008:
Strategy Tester Report for SVM_v_05_EURUSD EA

Could you or BARNIX provide this EA for download?

 

gimme gimme gimme