QD is double now. All Q_i[i] are replaced by QD[i]

to avoid numerical problems.

struct svm_model is moved to svm.h, so svm_destroy_model
is replaced svm_free_model_content and svm_free_and_destroy_model.

python_old is removed.

Makefile

@@ -1,6 +1,6 @@
 CXX ?= g++
 CFLAGS = -Wall -Wconversion -O3 -fPIC
-SHVER = 1
+SHVER = 2
 
 all: svm-train svm-predict svm-scale
 

Makefile.win

@@ -8,7 +8,7 @@ CXX = cl.exe
 CFLAGS = -nologo -O2 -EHsc -I. -D __WIN32__ -D _CRT_SECURE_NO_DEPRECATE
 TARGET = windows
 
-all: $(TARGET)\svm-train.exe $(TARGET)\svm-predict.exe $(TARGET)\svm-scale.exe $(TARGET)\svm-toy.exe
+all: $(TARGET)\svm-train.exe $(TARGET)\svm-predict.exe $(TARGET)\svm-scale.exe $(TARGET)\svm-toy.exe lib
 
 $(TARGET)\svm-predict.exe: svm.h svm-predict.c svm.obj
 	$(CXX) $(CFLAGS) svm-predict.c svm.obj -Fe$(TARGET)\svm-predict.exe

README

@@ -420,6 +420,80 @@ to classify new data.
     *NOTE* To avoid wrong parameters, svm_check_parameter() should be
     called before svm_train().
 
+    struct svm_model stores the model obtained from the training procedure.
+    It is not recommended to directly access entries in this structure.
+    Programmers should use the interface functions to get the values.
+
+	struct svm_model
+	{
+		struct svm_parameter param;	/* parameter */
+		int nr_class;		/* number of classes, = 2 in regression/one class svm */
+		int l;			/* total #SV */
+		struct svm_node **SV;		/* SVs (SV[l]) */
+		double **sv_coef;	/* coefficients for SVs in decision functions (sv_coef[k-1][l]) */
+		double *rho;		/* constants in decision functions (rho[k*(k-1)/2]) */
+		double *probA;		/* pairwise probability information */
+		double *probB;
+
+		/* for classification only */
+
+		int *label;		/* label of each class (label[k]) */
+		int *nSV;		/* number of SVs for each class (nSV[k]) */
+					/* nSV[0] + nSV[1] + ... + nSV[k-1] = l */
+		/* XXX */
+		int free_sv;		/* 1 if svm_model is created by svm_load_model*/
+					/* 0 if svm_model is created by svm_train */
+	};
+
+    param describes the parameters used to obtain the model.
+
+    nr_class is the number of classes. It is 2 for regression and one-class SVM.
+
+    l is the number of support vectors. SV and sv_coef are support
+    vectors and the corresponding coefficients, respectively. Assume there are
+    k classes. For data in class j, the corresponding sv_coef includes (k-1) y*alpha vectors,
+    where alpha's are solutions of the following two class problems:
+    1 vs j, 2 vs j, ..., j-1 vs j, j vs j+1, j vs j+2, ..., j vs k
+    and y=1 for the first j-1 vectors, while y=-1 for the remaining k-j 
+    vectors. For example, if there are 4 classes, sv_coef and SV are like:
+
+        +-+-+-+--------------------+
+        |1|1|1|                    |
+        |v|v|v|  SVs from class 1  |
+        |2|3|4|                    |
+        +-+-+-+--------------------+
+        |1|2|2|                    |
+        |v|v|v|  SVs from class 2  |
+        |2|3|4|                    |
+        +-+-+-+--------------------+
+        |1|2|3|                    |
+        |v|v|v|  SVs from class 3  |
+        |3|3|4|                    |
+        +-+-+-+--------------------+
+        |1|2|3|                    |
+        |v|v|v|  SVs from class 4  |
+        |4|4|4|                    |
+        +-+-+-+--------------------+
+
+    See svm_train() for an example of assigning values to sv_coef.
+
+    rho is the bias term (-b). probA and probB are parameters used in
+    probability outputs. If there are k classes, there are k*(k-1)/2
+    binary problems as well as rho, probA, and probB values. They are
+    aligned in the order of binary problems:
+    1 vs 2, 1 vs 3, ..., 1 vs k, 2 vs 3, ..., 2 vs k, ..., k-1 vs k.
+
+    label contains labels in the training data.
+
+    nSV is the number of support vectors in each class.
+
+    free_sv is a flag used to determine whether the space of SV should 
+    be released in free_model_content(struct svm_model*) and 
+    free_and_destroy_model(struct svm_model**). If the model is
+    generated by svm_train(), then SVs point to data in svm_problem
+    and should not be removed. For example, free_sv is 0 if svm_model
+    is created by svm_train, but is 0 if created by svm_load_model.
+
 - Function: double svm_predict(const struct svm_model *model,
                                const struct svm_node *x);
 
@@ -530,9 +604,15 @@ to classify new data.
     This function returns a pointer to the model read from the file,
     or a null pointer if the model could not be loaded.
 
-- Function: void svm_destroy_model(struct svm_model *model);
+- Function: void svm_free_model_content(struct svm_model *model_ptr);
+
+    This function frees the memory used by the entries in a model structure.
+
+- Function: void svm_free_and_destroy_model(struct svm_model **model_ptr_ptr);
 
-    This function frees the memory used by a model.
+    This function frees the memory used by a model and destroys the model
+    structure. This function is equivalent to svm_destroy_model, which
+    is deprecated after version 3.0.
 
 - Function: void svm_destroy_param(struct svm_parameter *param);
 

java/libsvm/svm.java

@@ -128,7 +128,7 @@ void swap_index(int i, int j)
 //
 abstract class QMatrix {
 	abstract float[] get_Q(int column, int len);
-	abstract float[] get_QD();
+	abstract double[] get_QD();
 	abstract void swap_index(int i, int j);
 };
 
@@ -143,7 +143,7 @@ abstract class Kernel extends QMatrix {
 	private final double coef0;
 
 	abstract float[] get_Q(int column, int len);
-	abstract float[] get_QD();
+	abstract double[] get_QD();
 
 	void swap_index(int i, int j)
 	{
@@ -309,7 +309,7 @@ class Solver {
 	byte[] alpha_status;	// LOWER_BOUND, UPPER_BOUND, FREE
 	double[] alpha;
 	QMatrix Q;
-	float[] QD;
+	double[] QD;
 	double eps;
 	double Cp,Cn;
 	double[] p;
@@ -501,7 +501,7 @@ void Solve(int l, QMatrix Q, double[] p_, byte[] y_,
 
 			if(y[i]!=y[j])
 			{
-				double quad_coef = Q_i[i]+Q_j[j]+2*Q_i[j];
+				double quad_coef = QD[i]+QD[j]+2*Q_i[j];
 				if (quad_coef <= 0)
 					quad_coef = 1e-12;
 				double delta = (-G[i]-G[j])/quad_coef;
@@ -544,7 +544,7 @@ void Solve(int l, QMatrix Q, double[] p_, byte[] y_,
 			}
 			else
 			{
-				double quad_coef = Q_i[i]+Q_j[j]-2*Q_i[j];
+				double quad_coef = QD[i]+QD[j]-2*Q_i[j];
 				if (quad_coef <= 0)
 					quad_coef = 1e-12;
 				double delta = (G[i]-G[j])/quad_coef;
@@ -707,7 +707,7 @@ int select_working_set(int[] working_set)
 					if (grad_diff > 0)
 					{
 						double obj_diff; 
-						double quad_coef=Q_i[i]+QD[j]-2.0*y[i]*Q_i[j];
+						double quad_coef = QD[i]+QD[j]-2.0*y[i]*Q_i[j];
 						if (quad_coef > 0)
 							obj_diff = -(grad_diff*grad_diff)/quad_coef;
 						else
@@ -731,7 +731,7 @@ int select_working_set(int[] working_set)
 					if (grad_diff > 0)
 					{
 						double obj_diff; 
-						double quad_coef=Q_i[i]+QD[j]+2.0*y[i]*Q_i[j];
+						double quad_coef = QD[i]+QD[j]+2.0*y[i]*Q_i[j];
 						if (quad_coef > 0)
 							obj_diff = -(grad_diff*grad_diff)/quad_coef;
 						else
@@ -953,7 +953,7 @@ int select_working_set(int[] working_set)
 					if (grad_diff > 0)
 					{
 						double obj_diff; 
-						double quad_coef = Q_ip[ip]+QD[j]-2*Q_ip[j];
+						double quad_coef = QD[ip]+QD[j]-2*Q_ip[j];
 						if (quad_coef > 0)
 							obj_diff = -(grad_diff*grad_diff)/quad_coef;
 						else
@@ -977,7 +977,7 @@ int select_working_set(int[] working_set)
 					if (grad_diff > 0)
 					{
 						double obj_diff; 
-						double quad_coef = Q_in[in]+QD[j]-2*Q_in[j];
+						double quad_coef = QD[in]+QD[j]-2*Q_in[j];
 						if (quad_coef > 0)
 							obj_diff = -(grad_diff*grad_diff)/quad_coef;
 						else
@@ -1135,16 +1135,16 @@ class SVC_Q extends Kernel
 {
 	private final byte[] y;
 	private final Cache cache;
-	private final float[] QD;
+	private final double[] QD;
 
 	SVC_Q(svm_problem prob, svm_parameter param, byte[] y_)
 	{
 		super(prob.l, prob.x, param);
 		y = (byte[])y_.clone();
 		cache = new Cache(prob.l,(long)(param.cache_size*(1<<20)));
-		QD = new float[prob.l];
+		QD = new double[prob.l];
 		for(int i=0;i<prob.l;i++)
-			QD[i]= (float)kernel_function(i,i);
+			QD[i] = kernel_function(i,i);
 	}
 
 	float[] get_Q(int i, int len)
@@ -1159,7 +1159,7 @@ float[] get_Q(int i, int len)
 		return data[0];
 	}
 
-	float[] get_QD()
+	double[] get_QD()
 	{
 		return QD;
 	}
@@ -1169,22 +1169,22 @@ void swap_index(int i, int j)
 		cache.swap_index(i,j);
 		super.swap_index(i,j);
 		do {byte _=y[i]; y[i]=y[j]; y[j]=_;} while(false);
-		do {float _=QD[i]; QD[i]=QD[j]; QD[j]=_;} while(false);
+		do {double _=QD[i]; QD[i]=QD[j]; QD[j]=_;} while(false);
 	}
 }
 
 class ONE_CLASS_Q extends Kernel
 {
 	private final Cache cache;
-	private final float[] QD;
+	private final double[] QD;
 
 	ONE_CLASS_Q(svm_problem prob, svm_parameter param)
 	{
 		super(prob.l, prob.x, param);
 		cache = new Cache(prob.l,(long)(param.cache_size*(1<<20)));
-		QD = new float[prob.l];
+		QD = new double[prob.l];
 		for(int i=0;i<prob.l;i++)
-			QD[i]= (float)kernel_function(i,i);
+			QD[i] = kernel_function(i,i);
 	}
 
 	float[] get_Q(int i, int len)
@@ -1199,7 +1199,7 @@ float[] get_Q(int i, int len)
 		return data[0];
 	}
 
-	float[] get_QD()
+	double[] get_QD()
 	{
 		return QD;
 	}
@@ -1208,7 +1208,7 @@ void swap_index(int i, int j)
 	{
 		cache.swap_index(i,j);
 		super.swap_index(i,j);
-		do {float _=QD[i]; QD[i]=QD[j]; QD[j]=_;} while(false);
+		do {double _=QD[i]; QD[i]=QD[j]; QD[j]=_;} while(false);
 	}
 }
 
@@ -1220,14 +1220,14 @@ class SVR_Q extends Kernel
 	private final int[] index;
 	private int next_buffer;
 	private float[][] buffer;
-	private final float[] QD;
+	private final double[] QD;
 
 	SVR_Q(svm_problem prob, svm_parameter param)
 	{
 		super(prob.l, prob.x, param);
 		l = prob.l;
 		cache = new Cache(l,(long)(param.cache_size*(1<<20)));
-		QD = new float[2*l];
+		QD = new double[2*l];
 		sign = new byte[2*l];
 		index = new int[2*l];
 		for(int k=0;k<l;k++)
@@ -1236,7 +1236,7 @@ class SVR_Q extends Kernel
 			sign[k+l] = -1;
 			index[k] = k;
 			index[k+l] = k;
-			QD[k] = (float)kernel_function(k,k);
+			QD[k] = kernel_function(k,k);
 			QD[k+l] = QD[k];
 		}
 		buffer = new float[2][2*l];
@@ -1247,7 +1247,7 @@ void swap_index(int i, int j)
 	{
 		do {byte _=sign[i]; sign[i]=sign[j]; sign[j]=_;} while(false);
 		do {int _=index[i]; index[i]=index[j]; index[j]=_;} while(false);
-		do {float _=QD[i]; QD[i]=QD[j]; QD[j]=_;} while(false);
+		do {double _=QD[i]; QD[i]=QD[j]; QD[j]=_;} while(false);
 	}
 
 	float[] get_Q(int i, int len)
@@ -1269,7 +1269,7 @@ float[] get_Q(int i, int len)
 		return buf;
 	}
 
-	float[] get_QD()
+	double[] get_QD()
 	{
 		return QD;
 	}
@@ -1311,7 +1311,7 @@ private static void solve_c_svc(svm_problem prob, svm_parameter param,
 		{
 			alpha[i] = 0;
 			minus_ones[i] = -1;
-			if(prob.y[i] > 0) y[i] = +1; else y[i]=-1;
+			if(prob.y[i] > 0) y[i] = +1; else y[i] = -1;
 		}
 
 		Solver s = new Solver();
@@ -1669,7 +1669,7 @@ private static void multiclass_probability(int k, double[][] r, double[] p)
 		int t,j;
 		int iter = 0, max_iter=Math.max(100,k);
 		double[][] Q=new double[k][k];
-		double[] Qp= new double[k];
+		double[] Qp=new double[k];
 		double pQp, eps=0.005/k;
 
 		for (t=0;t<k;t++)