mlp
所属分类:matlab编程
开发工具:matlab
文件大小:20KB
下载次数:33
上传日期:2010-12-01 04:32:31
上 传 者:
sabamadani
说明:
MLP with wieght and multilayer
文件列表:
mlp\MLP-APPROXIMATION-MATLAB\FileName1 (1869, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\mapweights.wts (623, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\MLP_Approximation.asv (20268, 2010-11-19)
mlp\MLP-APPROXIMATION-MATLAB\MLP_Approximation.m (20210, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\MLP_MapProcessing.asv (5160, 2010-11-19)
mlp\MLP-APPROXIMATION-MATLAB\MLP_MapProcessing.m (5052, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\outmap.txt (753, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\ProcessingResult.txt (1896, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\Twod.tra (6514, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB\Twod.tst (1869, 2010-11-22)
mlp\MLP-APPROXIMATION-MATLAB (0, 2010-11-22)
mlp (0, 2010-11-22)
This is a readme file for the MLP-Approximation source code written in MATLAB, available for free at:
http://www-ee.uta.edu/eeweb/ip/Software/Software.htm
NOTE: This is a MatLab 7.8 implementation of MLP-Approximation. The code have been designed on a Windows machine, but they should run on any Unix or Linux architecture with MatLab installed without any problems.
CONTENTS:
---------
I. Introduction
II. Getting Started
III. Basic Definitions
IV. Training Data Format
V. Included Training Data Files
I. Introduction
---------------
The software is meant to be a good starting point for people who like to develop their own code.
The source code written in MATLAB is essentially for training and processing a multilayer perceptron (MLP) neural
network.
1. MLP-Approximation.m - original source code for training a MLP network on regression/approximation (R/A) data.
The classic back propagation algorithm is used for training the MLP network.
II. Getting Started
-------------------
To use the source code to train a MLP network on R/A data, the first step is to obtain R/A type data.
The R/A type data should be in a standard format (explained below). Some R/A type data in the standard
format can be obtained from:
http://www-ee.uta.edu/eeweb/ip/training_data_files.htm
Once a R/A type data file is available in the standard format, compile and run the MLP-Approximation.c file.
Enter the necessary parameters, making sure that they match the R/A type training data provided as input for
the MLP.
Successful training results in displaying the training error in the command window and also saving the results and
the network weights as files, under the same directory as the source code.
Note: The saved weights are in a format that is compatible with our advanced software NuMap7. NuMap7 contains
advanced networks and training algorithms and a basic limited version is available for free download at:
http://www-ee.uta.edu/eeweb/ip/Software/Software.htm
Also, the saved networks can be pruned, tested and processed by NuMap7.
The saved network can then be used by the MLP-MapProcessing.c file along with the R/A type testing data file for
processing.
III. Basic Definitions
----------------------
For a given approximation or regression problem:
N - denotes the number of inputs in an input vector X.
M - denotes the number of outputs.
Training data file - a formatted file containing example data required for training a regression/approximation network.
Each pattern (row) in the file contains N input numbers, followed by the M desired outputs. The approximation type
training data file must have the data in the standard format for it to be used with this source code.
Pattern - a row vector in the training data file containing N inputs and M desired outputs. The dimension of a
pattern vector, also commonly called the number of elements per vector, therefore equals N+M.
IV. Training Data Format
------------------------
To comply with the standard format, the training data file:
- Should have the N Inputs to the left and the M Desired Outputs to the right.
- Should have data separated either by a spaces or tabs (space delimited or tab delimited) only.
- Should not have commas or other punctuation marks between data or between patterns.
- Should not have any column indices or row indices.
In other words, the training data file should purely consist of the inputs followed by the desired outputs for each
pattern. It should not have any other textual information.
Training data is required to form any regression/approximation network. Training data is nothing but raw data that
is formatted in a standard format to be used as input to the network. By standard format, we mean that each pattern
or feature vector (input) is followed by the corresponding desired output.
To make clear the structure of a training data file, let us consider the file for a simple MAX operator. The MAX
operator selects the maximum of all the inputs and sends it to the output. Below is a sample table of MAX operator
for 3 inputs and 1 output. Observe that the inputs are to the left and the desired (correct) output is to the right.
Inputs Desired Output
.5844768 .5359043 .6196933 .6196933
.1291312 .4173794 .3405759 .4173794
.0472856 .5994965 .5638752 .5994965
Each entry of 3 inputs and the corresponding output in the table constitutes a pattern. Thus there are 3 patterns
in the above table. The 3 inputs are in effect mapped to one output, which happens to be the largest of the 3 inputs.
Each input is separated by a blank space (space delimited data). It is important to realize that only space or tab
delimited data patterns work with this source code. Other formats, such as, comma delimited patterns or patterns with
column or row indices, will lead to erroneous results.
V. Included Training Data Files
-------------------------------
Training data file: TWOD.TRA
No. of Inputs: 8
No. of Outputs: 7
No. of Elements per vector: 15
No. of patterns: 1768
Testing data file present? YES
Testing data file: TWOD.TST
近期下载者:
相关文件:
收藏者: