MetaTrader and Neural Networks

New comment
 
Found at http://cortex.snowseed.com/forex_nn.htm
An article discussing:
1. How to create MTS that uses Neural Networks. What to use as inputs, how to avoid pitfalls.
2. How to export the (resulting) NN to MetaTrader script.

I have tried an example script they provided (at the very end of the article). It seems to be working. Then I have optimized it with MT optimizer. It seems to be working even better.
 

I have tried an example script they provided (at the very end of the article). It seems to be working. Then I have optimized it with MT optimizer. It seems to be working even better.


I agree:

Bars in test 17734
Ticks modelled 35368
Modelling quality n/a
Initial deposit 1000.00
Total net profit 4621.65
Gross profit 9363.95
Gross loss -4742.30
Profit factor 1.97
Expected payoff 47.65
Absolute drawdown 101.75
Maximal drawdown (%) 734.55 (14.0%)
Total trades 97
Short positions (won %) 38 (73.68%)
Long positions (won %) 59 (62.71%)
Profit trades (% of total) 65 (67.01%)
Loss trades (% of total) 32 (32.99%)
Largest
profit trade 521.25
loss trade -241.05
Average
profit trade 144.06
loss trade -148.20
Maximum
consecutive wins (profit in money) 12 (2018.00)
consecutive losses (loss in money) 4 (-334.30)
Maximal
consecutive profit (count of wins) 2018.00 (12)
consecutive loss (count of losses) -473.25 (2)
Average
consecutive wins 4
consecutive losses 2
 
I tried to follow the simple example (hopefully to learn more), and be able to build the script used above two times. The second time I let it learn for over 10 hours and the results were very poor compared to what the website posted (ofcourse), by using the same sample data and following directions exactly.
 
I tried to follow the simple example (hopefully to learn more), and be able to build the script used above two times. The second time I let it learn for over 10 hours and the results were very poor compared to what the website posted (ofcourse), by using the same sample data and following directions exactly.


I'd say you did something wrong. After all, I used the resulting script, as well as the script of my own.

Let's see.
0. What "simple example"? The final one is far from simple.
1. The data you used - where did you get them? I have used those that came with the program's archive, from samples\eurusd_h1.txt
2. The NN. It should learn 25000 epochs, not 10 hours. I don't know what computer you have, but it takes some time.
3. The results I have posted were obtained AFTER I have additionally optimized the expert. They are not the same as posted on site.
4. As far as I can tell, they have changed the version, so use the last build.
5. I have re-created the NN, by running the last script in their article, and the results are pretty much the same as they promised. Actually, I have published MY results, of MY network.
 
I copied and tried to run these files, but I'm getting an error message in the journal that it can't open the expert file.. I named it the same name as the examples.. anyone have any ideas..??

thanks in advance

david
 
I copied and tried to run these files, but I'm getting an error message in the journal that it can't open the expert file.. I named it the same name as the examples.. anyone have any ideas..??

thanks in advance

david



I FIGURED IT OUT.. thanks

d
 
Reply to Quark 22.09.05 23:01:
I am not a exactly a beginner, I have tons of programming experince and university degree in Computer Science and Engineering (some in AI too). But so what ;)

Simple example (not really simple) is the ONLY cortex tutorial http://cortex.snowseed.com/cortex.htm . The epochs that I ran had 2 or more digits after 25,000. Also I could not save the NN because I have not registered this software.
I am using the build of yesterday September 21, 2005 ver 5.0 (all my data (or other) files comes from this zip).

I have tested the expert like Quark and Nog mentioned with the following bad results: (with data file that came in zip from cortex).
__________________________________________________________________
Strategy Tester Report
Forex_Nn_Expert

Symbol EURUSD (Euro vs US Dollar)
Period 1 Hour (H1) 2003.02.06 02:00 - 2004.09.17 22:00
Model Every tick (based on all available least timeframes with fractal interpolation of every tick)
Bars in test 10136 Ticks modelled 1388444 Modelling quality 25.00%
Initial deposit 10000.00
Total net profit 0.00 Gross profit 0.00 Gross loss 0.00
Profit factor Expected payoff 0.00
Absolute drawdown 10000.00 Maximal drawdown (%) 0.00 (0.0%)
Total trades 0 Short positions (won %) 0 (0.00%) Long positions (won %) 0 (0.00%)
Profit trades (% of total) 0 (0.00%) Loss trades (% of total) 0 (0.00%)
Largest profit trade 0.00 loss trade 0.00
Average profit trade 0.00 loss trade 0.00
Maximum consecutive wins (profit in money) 0 (0.00) consecutive losses (loss in money) 0 (0.00)
Maximal consecutive profit (count of wins) 0.00 (0) consecutive loss (count of losses) 0.00 (0)
Average consecutive wins 0 consecutive losses 0
________________________________________________________________________

Quark: Did you register? How do you run script in cortex? meaning: I like to recreate the NN but don't know how or may have to register. Maybe you could put a link to where, how you recreate the NN and your results. I'd appreciate it.
 
I FIGURED IT OUT.. thanks


Hi David... I also figured that part out, but can't seem to get my expert working. It does not excecute any trades. Have you had a similar probem (before I start to delve deeper...)?
 

Simple example (not really simple) is the ONLY cortex tutorial http://cortex.snowseed.com/cortex.htm . The epochs that I ran had 2 or more digits after 25,000. Also I could not save the NN because I have not registered this software.
I am using the build of yesterday September 21, 2005 ver 5.0 (all my data (or other) files comes from this zip).


Aha! I mean.. well... :) As I mentioned, they have changed the build. Since then, some old "simple" examples are not working as expected. So what? They were never intended as "profitable" examples, just as "how to".

Also, they have added the http://cortex.snowseed.com/forex_nn.htm - the tutorial on using NNs specifically on Forex. I was ONLY referring to it.


Quark: Did you register? How do you run script in cortex? meaning: I like to recreate the NN but don't know how or may have to register. Maybe you could put a link to where, how you recreate the NN and your results. I'd appreciate it.


The "register" link is at the bottom of "cortex" page. ecommerce.snowseed.com/register.htm
I have registered. But the point is, you can use the same approach in your own soft, if you dislike Cortex (though I like it so far). First, create the NN. Second, export it as MQL script, no DLLs involved from that point on :)
 
Quark said: But the point is, you can use the same approach in your own soft, if you dislike Cortex (though I like it so far). First, create the NN.

I say that is what I was talking about from the beginning. The how to example is not working plus one can not save the NN without registering.
I just posted the expert results above in my post as I said I would. I have read all the cortex website pages (and that is saying a lot), because there is a lot to read. And by the way this is not my first time looking at this site. I have seen it develope and grow in the last few months.

Concolusion: There is a little bit of good information on this site, not worth the time spent reading all the site, some of which is confusing. Nothing on the site works as of this date. More development and debugging is needed.
I am going to spend my time looking over free open source (GNU) AI Neural Network software on web which is actually written more robust and with clear directions as what/how to use it.
 
I FIGURED IT OUT.. thanks


Hi David... I also figured that part out, but can't seem to get my expert working. It does not excecute any trades. Have you had a similar probem (before I start to delve deeper...)?


Once I figured out the naming problem with the icustom() function call my expert ran fine.. I placed trades and was profitable for as long as I let it run. the only way it would run was tick by tick which takes a very very long time to finish so I ended it before it finished..
123
New comment