matlab_neat
所属分类:人工智能/神经网络/深度学习
开发工具:matlab
文件大小:71KB
下载次数:43
上传日期:2012-03-06 15:03:02
上 传 者:
simple498
说明: 一类新的神经进化算法——NEAT,可同步进化BP网络的拓扑结构和连接权值,避免了网络结构的选择和设计,此程序是用于解决经典的XOR分类问题
(A new neuroevolution algorithm is called NEAT.NEAT can evolve the topologies as well as the connection weights of BP network. this matlab program solves the XOR problem.)
文件列表:
matlab_neat\initial_population.m (3675, 2003-05-19)
matlab_neat\neatsave.mat (55175, 2008-07-18)
matlab_neat\neat_main.m (15156, 2003-08-22)
matlab_neat\rep.m (1162, 1994-04-18)
matlab_neat\reproduce.m (30437, 2003-08-18)
matlab_neat\xor_experiment.m (4904, 2003-08-22)
matlab_neat\xor_experiment_vec.m (4136, 2003-08-22)
matlab_neat (0, 2010-12-21)
This distribution (v1.0, 8/31/03) contains a Matlab implementation of the
NeuroEvolution of Augmenting Topologies (NEAT) method for evolving
neural network topologies and weights. The package includes
an implementation of the XOR experiment as an example.
-The complete source code in this distribution was written by Christian Mayr,
who can be reached at: matlab_neat@web.de
-The NEAT method was developed by Kenneth Stanley and Risto Miikkulainen.
Additional information and papers describing NEAT in detail can be found
at Ken's website: http://www.cs.utexas.edu/users/kstanley/
A FAQ for Matlab NEAT may also become available through:
http://www.cs.utexas.edu/users/kstanley/neat.html
This program is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License version 2 as published
by the Free Software Foundation. This program is distributed in the hope
that it will be useful, but without any warranty; without even the implied
warranty of merchantability or fitness for a particular purpose. See the GNU
General Public License for more details.
We hope that this software will be a useful starting point for your own
explorations in neuroevolution. The software is provided as is,
however, we will do our best to maintain it and accommodate
suggestions. If you want to be notified of future releases of the
software or have questions, comments, bug reports or suggestions
related to the software or source code itself, send
email to matlab_neat@web.de. For general questions about NEAT,
send e-mail to kstanley@cs.utexas.edu, or refer to the NEAT
User's Page and FAQ at http://www.cs.utexas.edu/users/kstanley/neat.html.
NOTES:
-To run: After including the directory in which you installed matlab neat
in your startup.m file, just type "neat_main" at the Matlab
command line. The XOR experiment will run and graphs will be
displayed to indicate progress.
-Data Output: At the end of evolution, all population structures are in
memory. For example, to access the connection genes of population
member 12, you type at the prompt population(12).connectiongenes.
The entire population is stored in the population structure.
Matlab NEAT also outputs its structures in the file "neatsave.mat,"
which can be loaded in using "load 'neatsave'" to restore data from
a prior run. Please note that .mat files saved from Matlab NEAT
will only be usable on the same platform on which they were saved,
since they are saved in a binary format.
-Modifying the code: It should be possible to modify the code and
add experiments by using the extensive commenting as an aid
in understanding the system.
-XOR Termination Criterion: XOR is considered solved if the rounded
outputs are all correct. In other words, anything below
0.5 is round down to 0, and anything equal or above 0.5 is rounded
to 1. In the case the problem is solved, Matlab NEAT automatically
assigns a fitness of 16 in order to terminate.
-Vectorized XOR: Since the vectorized xor evaluation is about 3-5 times
faster (especially with large, densely connected networks), users
should use this version as a template. It may look a little strange
(at least for users not familiar with matlab), but the speed increase
justifies a little more coding effort. Those more comfortable
with the familiar can, of course, use the old (looped) version.
近期下载者:
相关文件:
收藏者: