1. How to prepare the data?
	(1) Download the file from website of LIBSVM data sets.
		http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets

2. How to compile the code (We use Method2 as an example)?
	(1) cd ./Method2
	(2) type ``make''

3. Usage
   (1) For Method1, Method2, Method2-con, Method2-g, and Subsampled, you can use ``./train -s 2 -x 10 -y 50 -z 20 data_name''
	[1] -s: solver name. For exmaple, ``-s 2'' is using LR.
	[2] -x: max ``outer'' iteration.
	[3] -y: max number of ``CG iteration''.
	[4] -z: number of split of training data set for subsampling Hessian. For example, ``-z 20'' is using 1/20 (5%) data to derive subsampled Hessian. 
   (2) For Method2-sg, you can use ``./train -s 2 -x 10 -y 50 -g 2 -z 20 data_name''
	[1] -g: number of split of training data to derive sampled gradient. For example, ``-g 2'' is using 1/2 (50%) data to derive sampled gradient.
		Notice that ``-g'' cooperate with ``-z''. For example, ``-g 2 -z 20'' means that using 1/2 (50%) data to get sampled gradient and then using 1/10 (0.5*0.1 = 5%) data
		which is from the 1/2 (50%) data using for gradient to get the subsampled Hessian.
   (3) For *-maxent, you can use ``./train -s 2 -x 10 -y 50 -z 20 train_data_name test_data_name''
	[1] Notice that we also provide the testing accuracy.
