add hack to disable optimization of linear svms; improved precision of optimize_linear_svm; add the relevant test, which however requires some big database (so it's disabled by default)

modules/ml/src/svm.cpp

@@ -1551,6 +1551,8 @@ void CvSVM::optimize_linear_svm()
         return;
 
     int var_count = get_var_count();
+    cv::AutoBuffer<double> vbuf;
+    double* v = vbuf;
     int sample_size = (int)(var_count*sizeof(sv[0][0]));
     float** new_sv = (float**)cvMemStorageAlloc(storage, df_count*sizeof(new_sv[0]));
 
@@ -1558,15 +1560,17 @@ void CvSVM::optimize_linear_svm()
     {
         new_sv[i] = (float*)cvMemStorageAlloc(storage, sample_size);
         float* dst = new_sv[i];
-        memset(dst, 0, sample_size);
+        memset(v, 0, var_count*sizeof(v[0]));
         int j, k, sv_count = df[i].sv_count;
         for( j = 0; j < sv_count; j++ )
         {
             const float* src = class_count > 1 && df[i].sv_index ? sv[df[i].sv_index[j]] : sv[j];
             double a = df[i].alpha[j];
             for( k = 0; k < var_count; k++ )
-                dst[k] = (float)(dst[k] + src[k]*a);
+                v[k] += src[k]*a;
         }
+        for( k = 0; k < var_count; k++ )
+            dst[k] = (float)v[k];
         df[i].sv_count = 1;
         df[i].alpha[0] = 1.;
         if( class_count > 1 && df[i].sv_index )
@@ -2570,7 +2574,8 @@ void CvSVM::read( CvFileStorage* fs, CvFileNode* svm_node )
         CV_NEXT_SEQ_ELEM( df_node->data.seq->elem_size, reader );
     }
 
-    optimize_linear_svm();
+    if( cvReadIntByName(fs, svm_node, "optimize_linear", 1) != 0 )
+        optimize_linear_svm();
     create_kernel();
 
     __END__;

modules/ml/test/test_save_load.cpp

@@ -133,4 +133,32 @@ TEST(ML_Boost, save_load) { CV_SLMLTest test( CV_BOOST ); test.safe_run(); }
 TEST(ML_RTrees, save_load) { CV_SLMLTest test( CV_RTREES ); test.safe_run(); }
 TEST(ML_ERTrees, save_load) { CV_SLMLTest test( CV_ERTREES ); test.safe_run(); }
 
+
+TEST(DISABLED_ML_SVM, linear_save_load)
+{
+    CvSVM svm1, svm2, svm3;
+    svm1.load("SVM45_X_38-1.xml");
+    svm2.load("SVM45_X_38-2.xml");
+    string tname = tempfile("a.xml");
+    svm2.save(tname.c_str());
+    svm3.load(tname.c_str());
+
+    ASSERT_EQ(svm1.get_var_count(), svm2.get_var_count());
+    ASSERT_EQ(svm1.get_var_count(), svm3.get_var_count());
+
+    int m = 10000, n = svm1.get_var_count();
+    Mat samples(m, n, CV_32F), r1, r2, r3;
+    randu(samples, 0., 1.);
+
+    svm1.predict(samples, r1);
+    svm2.predict(samples, r2);
+    svm3.predict(samples, r3);
+
+    double eps = 1e-4;
+    EXPECT_LE(norm(r1, r2, NORM_INF), eps);
+    EXPECT_LE(norm(r1, r3, NORM_INF), eps);
+
+    remove(tname.c_str());
+}
+
 /* End of file. */