Better NN EA development - page 21

[barnix]

Better EA structure:

[bt062]

i dont understand about nn but questioning

ok! i know nothing about nn nor the program, but still question

can i make like these

1st HL >< 2nd OC

1st OC >< 2nd HL

i used block as an input not tic

i want make it became a channel to filter out the noise

than to predict the future by channeling [road plan]

so when the price go as the road plan is save

but when it go out from the plan the got to call "huston we got situation here"

it we cek the situation as a sign

so if someone testing the market by a tic but the price follow the road plan it must be save

also we collecting the outsider tic for the possiblity for the next move, if to many high vilocity noise it mean we got something real thing

[barnix]

Indicators fo SVM outputs 1:

[David Melikisraelian]

neuro_ga

barnix:
#_LinRegres_Neuro_GA.mq4

Thanks

Barnix

I have been looking at the code, it seems to be working good, has anyone used it before, if so what do you think about it.

[David Melikisraelian]

Pnn

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)

Model Every tick (the most precise method based on all available least timeframes)

Parameters Lots=0.2; pBalanceTraded=1; adjustLotsOk=true; tradeAtDayClose=true; sl=140; tp=10; rp=0.7; pnn="----------pnn parameters----------"; sigma=0.1; fcastbars=2; Patterns=30; Inputs=6;

Bars in test 1053 Ticks modelled 505698 Modelling quality 90.00%

Mismatched charts errors 66

Initial deposit 1000.00

Total net profit 292.34 Gross profit 328.60 Gross loss -36.26

Profit factor 9.06 Expected payoff 18.27

Absolute drawdown 8.60 Maximal drawdown 68.01 (6.05%) Relative drawdown 6.05% (68.01)

Total trades 16 Short positions (won %) 11 (90.91%) Long positions (won %) 5 (80.00%)

Profit trades (% of total) 14 (87.50%) Loss trades (% of total) 2 (12.50%)

Largest profit trade 43.58 loss trade -31.58

Average profit trade 23.47 loss trade -18.13

Maximum consecutive wins (profit in money) 11 (217.01) consecutive losses (loss in money) 1 (-31.58)

Maximal consecutive profit (count of wins) 217.01 (11) consecutive loss (count of losses) -31.58 (1)

Average consecutive wins 7 consecutive losses 1

# Time Type Order Size Price S / L T / P Profit Balance

1 2008.03.17 00:00 sell 1 0.20 2.0191 2.0339 0.0000

2 2008.03.18 00:00 close 1 0.20 1.9972 2.0339 0.0000 43.48 1043.48

3 2008.03.20 00:00 buy 2 0.21 1.9836 1.9691 0.0000

4 2008.03.25 00:00 modify 2 0.21 1.9836 1.9843 0.0000

5 2008.03.26 00:02 close 2 0.21 2.0020 1.9843 0.0000 39.87 1083.35

6 2008.03.28 00:00 sell 3 0.22 2.0067 2.0206 0.0000

7 2008.03.31 00:00 close 3 0.22 1.9937 2.0206 0.0000 28.25 1111.59

8 2008.03.31 00:00 buy 4 0.22 1.9937 1.9792 0.0000

9 2008.04.01 05:40 s/l 4 0.22 1.9792 1.9792 0.0000 -31.58 1080.02

10 2008.04.03 00:00 buy 5 0.22 1.9869 1.9726 0.0000

11 2008.04.04 00:00 close 5 0.22 1.9938 1.9726 0.0000 15.50 1095.52

12 2008.04.07 00:00 sell 6 0.22 1.9930 2.0072 0.0000

13 2008.04.08 00:00 close 6 0.22 1.9880 2.0072 0.0000 10.65 1106.16

14 2008.04.10 00:00 sell 7 0.22 1.9733 1.9874 0.0000

15 2008.04.14 00:01 close 7 0.22 1.9675 1.9874 0.0000 12.06 1118.22

16 2008.04.14 00:01 buy 8 0.22 1.9675 1.9530 0.0000

17 2008.04.15 00:00 close 8 0.22 1.9727 1.9530 0.0000 11.76 1129.98

18 2008.04.17 00:00 buy 9 0.23 1.9718 1.9575 0.0000

19 2008.04.18 00:00 close 9 0.23 1.9906 1.9575 0.0000 43.58 1173.56

20 2008.04.21 00:00 sell 10 0.23 1.9981 2.0122 0.0000

21 2008.04.22 00:00 close 10 0.23 1.9827 2.0122 0.0000 35.05 1208.61

22 2008.04.24 00:00 sell 11 0.24 1.9788 1.9929 0.0000

23 2008.04.25 00:00 close 11 0.24 1.9728 1.9929 0.0000 14.02 1222.63

24 2008.04.28 00:00 sell 12 0.24 1.9831 1.9972 0.0000

25 2008.04.30 00:00 close 12 0.24 1.9682 1.9972 0.0000 34.99 1257.62

26 2008.05.01 00:00 sell 13 0.25 1.9867 2.0006 0.0000

27 2008.05.02 00:00 close 13 0.25 1.9781 2.0006 0.0000 21.10 1278.72

28 2008.05.05 00:00 sell 14 0.26 1.9738 1.9877 0.0000

29 2008.05.07 00:00 close 14 0.26 1.9715 1.9877 0.0000 5.15 1283.87

30 2008.05.08 00:00 sell 15 0.26 1.9507 1.9646 0.0000

31 2008.05.14 00:00 close 15 0.26 1.9450 1.9646 0.0000 13.16 1297.02

32 2008.05.15 00:00 sell 16 0.26 1.9461 1.9602 0.0000

33 2008.05.15 23:59 close at stop 16 0.26 1.9479 1.9602 0.0000 -4.68 1292.34

[superluz]

[barnix]

Intermarket Analyzis with SVM

[superluz]

[peri]

[lvsefa]

superluz:
Hi Barnix,

I could run the batch file with no problem (see pic)

I’m testing only with fw 10

My batch:

rem c=128.0 g=2.0

call C:\Mia\Barnix\svmlib3\run.bat 128.0 2.0 EURUSD.libsvm C:\Mia\Mig1\experts\files\b4\dd 0 338 C:\Mia\Barnix\svmlib3 C:\Mia\Mig1\experts\files\b4\dd C:\S\B4\DD

The only problem I get is that the “#SVM_b4_dd_pred0_” return always :

2008.05.20 10:04:20 #SVM_b4_dd_pred0_ EURUSD,M5: ddraw 2147483647.00000000

2008.05.20 10:04:20 #SVM_b4_dd_pred0_ EURUSD,M5: ddraw 2147483647.00000000

2008.05.20 10:04:20 #SVM_b4_dd_pred0_ EURUSD,M5: ddraw 2147483647.00000000

2008.05.20 10:04:20 #SVM_b4_dd_pred0 EURUSD,M5: b4/dd/pred0/EURUSD.libsvm.t-> 1211169600 1211169900

2008.05.20 10:04:20 #SVM_b4_dd_pred0 EURUSD,M5: b4/dd/pred0/EURUSD.libsvm.prd-> 3 3

2008.05.20 10:04:20 #SVM_b4_dd_pred0 EURUSD,M5: b4/dd/pred0/EURUSD.libsvm.t-> 1211169600 1211169900

Any idea what I’m doing wrong?

Thanks again for the exceptional work that you did !!

In my opinion, in barnix's pics he use 7 classes to make #SVM_b4_dd_pred indicator set, but the batched code file has only 3 classes, so I think you should do more work to make it perfect to achieve barnix's results.

And in barnix's early post's attached files, he showed some svm predict file with 1,2....7 predict results, in you pic seems that it got only one predict class as result,I think the test set used or results has something wrong...