Better NN EA development - page 58
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Thanks for the fast reply.
My terminal crashes when i try to apply attrib indicators, i put everything where you said.
I will try on a fresh mt4 install and will tell you if works or not.
use winxp ver 2002 sp 2
Thanks for the fast reply. My terminal crashes when i try to apply attrib indicators, i put everything where you said.
That what i use, windows xp service pack 2. I have several notebooks on vista, and DUH !
Files we need: just the mqh file, the library dll and the indics ? Do we need more ?
Oh Barnix, excuse me it works
.
I made a fresh mt4 install on my laptop (vista sick !)
And it works.
Tomorrow i will try on xp.
Thanks, and sorry.
I was impressed by the advertsement of Fapturbo. but, look into their so called real statement, it was obviously a demo acc(they claim it's a real). I wonder they only shew you the good, hided the blow up. it's too good to be true
Barnix,
I see that you've tested FAPturbo.
1) Do you test the scalper or the strategier?
2) On which time frame? and broker?
I tried to backtest it, and could not make any trade in the backtest. So I thought that the developer has disabled the backtest function for the FAPturbo. Apparently, it isn't so. You have proved that the FAPturbo can be backtested.OFFICIAL NEWS :
The first july 2009, microsoft will give a coupon to vista users to install windows seven, the successor of vista.
Man, I wish I was smart enough to comprehend all the stuff that Barnix is posting. However, even though I don't necessarily comprehend the content I can understand the massive work he is putting in and the selfless way he is sharing information.
So I'd like to say thank you to Barnix, this is very impressive. I am not sure if this is just a function of me being new but is there a way I can PM you Barnix?
Great thread keep up the work!
Ben
Nn
Hi Dear Mr Barnix,
thx for new posts ,
these are the result of my NN developments ,
a. network setups : ( variable input neurons num , 3 layer (outputs are linear ) , sigmoid function for hidden layer neurons , variable output neurons num.
1, 2, .... 120, candle datas as input ( any combination of TOHLCV ) ( max input nodes 120*6 + some consts )
, 1,2, .... 8 candles output ( any combination of TOHLCV ) as output ( max out neurons 8*6 + some feedback outputs ) for : one or more CPairs and one or more TFs ( eg. single block TF1D EURUSD or colony of TF5m-TF15m-TF30m-TF1H-TF4H-TF1D-TF1W of EURUSD-USDJPY-GBPUSD-USDCAD-USDCHF-AUDUSD-NZDUSD-GOLD total=49 blocks of interconnected networks + some feedbacks ) and the largest one has 319 blocks of networks,
b. Training and testing data set ( 2 years of Alpari history of TF5m-TF15m-TF30m-TF1H-TF4H , 12years of Alpari history for TF1D-TF1W ) - all of those data are controlled for spikes and missed datas ( spikes are modified to nearest meaningful values and missed candles filled by last close value )
c. network blocks sizes ( in-h1n-h2n-out from 5-20-20-2 to 750-3000-3000-70 (nearly 11,000,000 synapses) )
d. training duration : 72 hours(more than 22,000,000 epochs for smallest one) , to 250 hours(less 20,000 epochs for largest one )
e. networks count for training: nearly 100 configuration ( I use 1 Quad Core PC and 2 Dual core PCs , each core for one of network trainings )
f. Ram usage of training : ( less than 20MB for smallest one and more than 1.6GB for largest one )
g. the result of testing after 7months working ...
best : EURUSD daily HL prediction for next day +- 35 PIPs error
EURUSD daily Direction prediction for next day better than %80
good : EURUSD 15m HL prediction for next 3 candles +- 10 PIPs error ( it seems like predictive for MA(High) and MA(Low)
EURUSD 15m Direction prediction for next 3 candles better than %65
results of other CPAIRs prediction have lower accuracy and the least is USDJPY.
h. my next steps are ,
1-training to generate Trade/Hold/Exit Signal from only TOHLCV input data ( turning point detection ),
2-training to generate Trade/Hold/Exit Signal from Indicators combination (like SVM) and TOHLCV input data ( turning point detection ).
notification1, the best results achieved by medium size NN ( eg: 40-160-160-5 ) after 3,000,000 epochs : 100+hours training , the larger networks are too heavy for my PCs to continue training more epochs, then I stopped them ,
notification2, overfitting and memorizing are encountered when LR > 0.1 , my best LR <= 0.05 ( the larger the network must the lower LR )
notification3, Library that I have used, is : LWNN that I adressed it earlier (page 34) , and for CSV data file parsing I have used Some codes of Mr Barnix ( fx5 and fx10 ) thx alot ... , and at last I coded In C and C++ and Console application of MSVC.
notification4, I have combined 2 years of FUNDAMENTAL NEWS ( economic indexes eg, IR, CPI, TradeBalance ,NFP ,Crude oil, ... (more than 40 for every Country!) but my networks couldn't extract relations! or those FUNDAMENTALs didn't help the network to make good results.
thx .