Overview:

This is an educational version of PCNeuron, a neural network development shell developed by Professor I-Cheng Yeh at the Chung-Hwa Institute, Taiwan in 1992. The PCN system was written in Turbo C (version 2.0) and can be run in a PC but only under the DOS Mode.

PCN coverage is quite comprehensive. It can be used to build, train, and test 11 different neural networks, including perceptron (Per), backpropagation (BPN), probabilistic neural network (PNN), learning vector quantization (LVQ), counter-propagation (CPN), self-organizing map (SOM), adaptive resonance theory (ART) network, Hopfield neural network (HNN), Hopfield-Tank neural (HTN) network, and annealed neural network (ANN). The educational version restricts users to develop neural networks with up to 50 neurons in each layer. The professional version, however, extends its capacity up to machine's available RAM and includes two more networks - improved backpropagation (IBP) and neural expert systems (NES).

[ Back to Top of the Page ]

Information Flow of PCN:

The development of neural network involves the following key steps:

1. Determine the input and output variables to be considered.
2. Select an appropriate neural network model.
3. Represent the problem in terms of the selected neural network model.
4. Collect data.
5. Preprocessing the data (scaling, transferring and filtering).
6. Divide the data into training and testing data sets.
7. Train and test the result.
8. Postprocessing the result (plotting, analyzing, etc.).
9. Use the trained weight data for prediction (most public domain system don’t have this option).

The design of PCN follows the above generic procedure. Figure 1 shows the specific information flow of PCN:

Figure 1: Information Flow of PCN

[ Back to Top of the Page ]

Menu Hierarchy of PCN:

PCN is a menu-driven system. Table 1 shows its menu hierarchy. The first hierarchy contains five modules: file, pre, build, run, and post:

1. File. This module allows users to load and save files, jump to DOS environment, and exit the system.
2. Pre. This module provides all mathematical functions needed to preprocess data for variables. Variable Statistic can be used to compute maximum, minimum, mean, standard deviation, and perform probability scaling. Example Process & Sampling can be used to divide data into training and testing data sets according to the type of data user has.
3. Build. This module is provided to setup and modify parameter file. The user needs to choose the type of neural network to be developed first, the system will then list the needed parameters that matched with the selected network type for entry or revision.
4. Run. Once the parameter file was built, user can then start to train and test the network by selecting this menu. The training and testing results will then be displayed in screen for user to review and saved in a file for printing. If the results is not as good as user expected, s/he can go back to modify the structure or parameter values.
5. Post. This module provides seven popular analytical methods - convergence curve, scatter diagram, confusion matrix, ROC curve, SOM matrix, weight histogram, and rescaling for user to evaluate the performance of the network s/he developed.

[ Back to Top of the Page ]

Files Used by PCN:

Input Files:

• net: file used to store parameters.
• tra: file used to store training data set.
• tes: file used to store testing data set.

• res: file used to store results.
• cur: file used to store data related to the progress of convergence.
• wei: file used to store connecting weight.

• ori: file used to store original data.
• fin: file used to store final scaled data.
• sta: file used to store statistical analysis results for data.
• lst: list of parameter values in text format.

[ Back to Top of the Page ]

System Installation:

PCN was stored in my course server under directory "systems\pcn" in a compressed format to save disk space. The following steps can be used to install the system:

1. Create a directory (e.g., PCN) in the hard drive and download the two files - "exe.exe" and "exam.exe" into that directory.
2. Run "exe.exe" and "exam.exe" from Windows Explorer (by click on the fil name). The program extract all the necessary files automatically.

[ Back to Top of the Page ]

Run the System: