1. How to produce the data for this tool?
	(1) Download the file from website of LIBSVM data sets (training and testing data sets or only training data) and put it into the directory "./data" 
		http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass.html

	(2) Scale the data to [0,1] if the data set is not be scaled. Use the tools in LIBSVM.
		./libsvm-3.12/svm-scale -l 0 -u 1 your_data > your_data_scale
	    Note that: If there are training data set and testing data set,
	    you can use the following command:
	    [1] cat data data.t > data.total
	    [2] ./libsvm-3.12/svm-scale -l 0 -u 1 data.total > data.totla.scale
	    [3] head -n #training data.total.scale > data.scale
	    [4] tail -n #testing data.total.scale > data.scale.t

	(3) Using the following code to produce the format for MATLAB
	    [1] cd ./libsvm-3.12/matlab
	    [2] Type ``make'' in order to make libsvmread and libsvmwrite
	    [3] matlab -r "trans_HF('your_data_scale',20)" where ``20'' means that splitting trainging data into 20 parts. 
	
   Finally, you can see the file "your_data_scale.mat" in the ./mat

2. How to run this code? (This is the example for running Comb2. For others, they are the same.)
	(1) cd ./Comb2
	(2) matlab -r "nnet_demo_2('your_data_scale',1,1,20,2/3,3/2)" 
		[1] filename
		[2] damping flag
		[3] sampling flag
		[4] number of splits of training data set for sampling Hessian
		[5] drop constant for LM method
		[6] boost constant for LM method
	(3) The log file will be put in the Comb2 directory and its name is ``your_data_scale.txt''.
