[med-svn] r7025 - trunk/packages/rostlab/profnet/trunk/debian

[Steffen Möller]

Author: moeller
Date: 2011-06-15 13:40:46 +0000 (Wed, 15 Jun 2011)
New Revision: 7025

Modified:
   trunk/packages/rostlab/profnet/trunk/debian/README.Debian
Log:
Helping the readme with a typo and a restructured last setence.


Modified: trunk/packages/rostlab/profnet/trunk/debian/README.Debian
===================================================================
--- trunk/packages/rostlab/profnet/trunk/debian/README.Debian	2011-06-15 13:37:08 UTC (rev 7024)
+++ trunk/packages/rostlab/profnet/trunk/debian/README.Debian	2011-06-15 13:40:46 UTC (rev 7025)
@@ -1,9 +1,22 @@
 profnet
 =======
-profnet_* binaries are neural network implementations in Fortran.  Due to the original design of the code, a specific binary is compiled for each particular network architecture, changing certain constants in the source code.  Therefore, there is a binary for every network architecture used.  Note: certain array structures are intentionally indexed out of bounds in some of the binaries.
+profnet_* binaries are neural network implementations in Fortran.
+Due to the original design of the code, a specific binary is compiled
+for each particular network architecture, changing certain constants
+in the source code.  Therefore, there is a binary for every network
+architecture used.  Note: certain array structures are intentionally
+indexed out of bounds in some of the binaries.
 
 Q: Why so many binary packages?
 -------------------------------
-There are a handful of prediction methods built around each of the binary profnet packages.  Each depends on the mathching profnet binary package.  Each prediction method requires a different neural network architecture and therefore - due to the design of the code - a different binary.  Binaries are compiled with constants set to the architecture of the network and are therefore not reusable for other architectures.  Improving this code is not planned.  A reimplementation of these networks with a neural network library is underway.
+There are a handful of prediction methods built around each of the binary
+profnet packages.  Each depends on the matching profnet binary package.
+Each prediction method requires a different neural network architecture
+and therefore - due to the design of the code - a different binary.
+Binaries are compiled with constants set to the architecture of
+the network and are therefore not reusable for other architectures.
+To further develop the code beyond regular maintenance for compiler and
+architecture updates is not planned since a complete reimplementation
+of these networks with a neural network library is already underway.
 
  -- Laszlo Kajan <lkajan at rostlab.org>  Tue, 14 Jun 2011 18:50:52 +0200