read first.txt

来自「bp, backpropagation 神经网络 简单认字 错误开发实例。」· 文本 代码 · 共 40 行

For Backpropagation character recognition:

Must have microsoft office installed. Referenced Microsoft Excel Object Library (12.0) to aids debug.


Containing:
	one form 		- main.frm,
	two class file 		- classTraining.cls & classRecognize.cls
	one bas/module file	- basFormulae.bas
	one text file		- data.txt

basFormulae.bas:-
Containing formulae for sigmoid(x)/f(x) and revSigmoid(x)/f'(x) function
Can use sigmoid(x) and revSigmoid(x) functions in any form/class within project
I choose bipolar sigmoid as activation function, page 293, f2(x)

classTraining.cls
Contain all functions need during training, page 294 to page 296
Used in main.frm, by declaring 26 object of classTraining (Dim clsTrain(1 To TOTAL_TRAINING_SET) As ClassTraining)
Each object store informations of weights and etc, eg: clsTrain(1) use to train "A", so it contains all information calculated inside it. clsTrain(2) contains details of "B", and so on.

classRecognize.cls
Contains all function need for application/recognize, page 299 to 300.
Same case with classTraining.  

data.txt
containing input for training, from A to Z, one set for each character.
Data must be binary form, but when loading by the program, the data will convert to bipolar, by replacing 0 with -1. So the program is training and testing using bipolar values.


Other forms:
lab1Hebb.frm		-first lab assigment
lab2perceptron.frm	-second lab assigment
lab3perceptron.frm	-3rd lab assigment
These 3 forms have no relation with the recognition application.



A111083, A115216, A110681
Nov 8 '07