|
PureBytes Links
Trading Reference Links
|
At 13.01 11/02/03 -0500, you wrote:
Are
you using back propagation or some other kind of
NN?>Are
there any publicly available NN plug-ins for AB?
Can
you recommend any particular NN tools for beginners that will integrate
easily with AB?
Sure! I use time lagged recurrent network developed with
neurosolutions:
http://www.neurosolutions.com/
(it is not exactly a cheap software)
Neurosolutions provides "custom solution wizard" that, starting
from your network, builds a dll that you can use in other software.
I wrote a plugin for Amibroker that works with dlls generated with custom
solution wizard and the result function is very simple:
result=neural("path
of
dll","path
of weights
file",input1,input2, ...
,inputN);
that recall network provided by the dll, uses the weigth file you specify
and takes, as input, the input array input1, ... , inputN
In order to make the development process more easy, in wrote
another function that lets you export data in ascii files easly:
for example:
io=(O-Ref(O,-1))/Ref(O,-1);
ih=(H-Ref(H,-1))/Ref(H,-1);
il=(L-Ref(L,-1))/Ref(L,-1);
ic=(C-Ref(C,-1))/Ref(C,-1);
export("D:\\NeuroSolutions
4\\MIEI\\BlackBox\\Dati\\"+FullName()+".txt",";",io,ih,il,ic,1);
exports, in an ascii file "called name_of_the_stock.txt", four
data array (io, ih, il, ic) with ";" as separator and the last
"1" tells to the function to export only the data relative to
bars in which *all* the input data are defined.
For example:
input1=MA(C,2);
input2=MA(C,10);
export(FullName()+".txt",";",input1,input2,1);
creates a file like this:
data;arg1;arg2
17/1/1997;0.22809544;0.22519891
20/1/1997;0.23009582;0.22650582
21/1/1997;0.22960240;0.22735131
22/1/1997;0.22858888;0.22837549
....
while:
input1=MA(C,2);
input2=MA(C,10);
export(FullName()+".txt",";",input1,input2);
creates this file:
data;arg1;arg2
2/1/1997;;
3/1/1997;;
7/1/1997;0.21884042;
8/1/1997;0.21924049;
9/1/1997;0.22213435;
10/1/1997;0.22529492;
13/1/1997;0.22776204;
14/1/1997;0.23006915;
15/1/1997;0.22916231;
16/1/1997;0.22782874;
17/1/1997;0.22809544;0.22519891
20/1/1997;0.23009582;0.22650582
21/1/1997;0.22960240;0.22735131
...
The export function simplify the process to develop network because you
can create input and desired files very fast.
network function recalls network, but not start learning process: I wrote
a function which train the network directly from AB but, since this is a
slow process, the plugin doesn't work correctly: AB stops to work.
So, if you want to train the network, you must use
neurosolutions.
At this moment, my plugin is not public available but, If you would like
to have it, write me.
hola
Buttha
Yahoo! Groups Sponsor
ADVERTISEMENT
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.
|