|
Home
Software
Yahoo Charts
NEW:
ADVANCED
TRADESTATION TECHNIQUE
US Markets Daily
snapshots
Technical
Description
| |
Neural Net Developments
We currently offer 3 neural network
software programs:
- NXL is a C language implementation of a neural net
which learning algorithm and structure are closely inspired from
Scott Fahlman's QuickProp. It has been written in C and compiled on the form of a DLL for
use with VBA, for instance Microsoft Excel, hence the name NXL. Sample spreadsheets
are provided with VBA
code.
- WinSOM is a Windows based Kohonen SOM developed
in C++. It offers 2D mapping capabilities, and is particularly easy to
use.
- VB BackProp is a small Visual Basic utility for neural net novices who are
familiar with Excel and VBA. The program differs from many existing
neural net utilities as it gives the user a lot of levers to better apprehend
the training process. While it can be used as is for small neural
modeling applications, it is recommended to acquire the VBA code, to better
customize it to your application environment.
Please note that VB BackProp will no longer be updated.
All emphasis is now put on our NXL library.
Notes on NXL (2005)
NXL has now been converted to NXL2 with a more detailed API. NXL2 has been
recompiled using Microsoft Visual C, which brings new API declarations.
The VB code remains unchanged, hence this will not affect your work. There
are a few additional functions and accessors to make your neural nets "lean and
mean". Many neural nets do not generalize well on account of "neuronal
glut", and it is always advisable to build smaller nets.
Please also note the following:
- There is a silly bug in 'Average Test Error' calculation in current NXL.
It will not be fixed in NXL, but has been fixed in NXL2. For users who
wish to stay with NXL, please use TotalTestError which is correct, then of
course divide it by the number of patterns in your set;
- The new API calls are listed here;
- All calculations have been changed to 'double' precision;
- A new user protection scheme is in place with NXL2.
New codes will
be needed for existing users. Upgrade is obviously free of charge for
existing registered users.
- new sample spreadsheets are available on the download page.
|
