readme

来自「- XCS for Dynamic Environments + Conti」· 代码 · 共 48 行

/*	XCSR_DE1.0
 *	--------------------------------------------------------
 *	Learning classifier system based on accuracy in dynamic environments
 *
 *	by  Huong Hai (Helen) Dam
 *	[e-mail]
 *	UNSW @ ADFA, Canberra Australia
 *	Artificial Life and Adaptive Robotics Laboratory
 *	http://www.itee.adfa.edu.au/~alar
 *
 *	Last modified: 24-11-2005
 *
 */
XCS is explored in dynamic environments with different magnitudes of  change to  the underlying concepts. We found that the conventional XCS is capable of
recovering quickly when dealing with small magnitudes of change. However, when the magnitude of change is high, it requires a long recovery time for the system to achieve a stable performance.

We proposed three strategies, which aim to reduce the recovery time of XCS after concepts change. We found that the re-initialized population strategy dramatically reduces recovery time in a noise free environment. The adaptive learning approach is the next best. When noise is incorporated into the training data, the
adaptive learning strategy achieves a better predictive performance when compared to other strategies.

To compile in unix:
g++ -o main *.cpp

To run:
10 input arguments:
cout<<"seed: random seed			"<<endl; 
cout<<"noise: noise level			"<<endl;
cout<<"population: maximum population size	"<<endl;
cout<<"crate: crossover rate			"<<endl;
cout<<"mrate: mutation rate			"<<endl;
cout<<"representation type: continuous-valued"<<endl;
cout<<"recovering concept drift strategy	"<<endl;
cout<<"a real threshold before changing		"<<endl;
cout<<"a real threshold after changing		"<<endl;

+Representation_type (for continous-valued representation): 
1 - Min max representation
2 - Centre spread representation
4 -  Min Percentage representation
+Recovering concept drift strategies
0 - Traditional XCS
1 - Reinitialized population after the change
2 - Reinitialized parameters after the change
3 - Adaptive learning rate according to the system performance

Ex>>./main 1000 0.0 1000 0.8 0.04 4 1 0.1 0.9

Related publications:
a.	H.H. Dam, H.A. Abbass, C.J. Lokan, Evolutionary Online Data Mining