Originally Posted by witchazel

its all based on williams accelerator/decelerator, which the formula is

AO = SMA(median price, 5)-SMA(median price, 34)
AC = AO-SMA(AO, 5)

here is the code in EA

extern int x1 = 135;
extern int x2 = 127;
extern int x3 = 16;
extern int x4 = 93;

double w1 = x1 - 100;
double w2 = x2 - 100;
double w3 = x3 - 100;
double w4 = x4 - 100;
double a1 = iAC(Symbol(), 0, 0);
double a2 = iAC(Symbol(), 0, 7);
double a3 = iAC(Symbol(), 0, 14);
double a4 = iAC(Symbol(), 0, 21);
return(x1* a1 + x2 * a2 + x3 * a3 + x4 * a4);

so it ends up being


return(35 * ac(this bar) + 27 * ac(7 bars ago) + -84 * ac(14 bars ago) ) + -7 * ac(21 bars ago) );

if it ends up being greater then 0 then it buys and if less then it sells. looks like its alway in the market as well. It only opens 1 order at a time though.

It looks like a variation of the alligator by Williams but it uses 4 ACs instead of 3.

how they pick the x1,x2,x3,x4 numbers? i dont know, seems random to me
why they -100 from each of them? no idea, should have just enterd a smaller number

Originally Posted by Craig

My 2 cents, Neural Networks don't really work as far as I can see, I have spent countless hours coding and testing such beasts with nothing really to show.
When evaluating such methods it is important to understand what is actually happening, to cut though the BS as it were. Neural Networks are not magic, just non-linear regression, as such they can do an amazing job of fitting test data but tell you nothing of the future. The other mistaken assumption which is applied to NN & Forex is that there is some type of 'hidden pattern' in the data which the NN will deduce, I have not found this to be the case. In my experence the greater the degrees of fredom of a system the more prone it is to overfitting the data, hence giving no value in data ranges outside the test data.

Originally Posted by Craig

I tweaked this EA to follow a more conventional NN layout (Normalized inputs, better activation functions, etc) and tried running optimizations over 1 years worth of data. At 90% quality it was modistly profitable, but performed poorly on out of sample data. Running the the same tests on tick data there was even less profit, but with better out of sample performance.

So in summary, it was just profitable ($2000 over the course of a year, starting with $10000).

Running the optimizations on Metatrader is kind of limited because the genetic optimizer freaks if you have too many variables to optimize (e.g. weights in a NN), I posted this bug on the Metaquotes site, but nobody appears to give a s**t.

Originally Posted by forex_for_life

I've never heard mention of this very interesting tool that has been improving already great trading systems for a while now, or so I read. Has anybody had any experience with this yet?

Lil info on em for those who do know. Seems to be very "revolutionary".

http://cortex.snowseed.com/index.html
(This site has some other interesting topics to help develop a trader's psychology. Plus they offer a free NN software.)

http://www.neuroshell.com/learnmore.asp
(These people sell their software. Lot of free info on this site. Plus they've gotten awards from Stocks and Commodities mag.)


P.S.

It looks like it helps to have an understanding of C++, which a lot of you do.

*NOTE*

I'm not selling, soliciting, offering, or marketing anything. Just thought this was very interesting.

Originally Posted by prasxz

NN makes more acurate to recognize price pattern through data mining ...
You can try this Indi based on NN EA

==================
Forex Indicators Collection