add cross validation option for R interface

Xtra-Computing · Jun 27, 2018 · ef7e7b6 · ef7e7b6
1 parent a114e8b
commit ef7e7b6
Show file tree

Hide file tree

Showing 3 changed files with 42 additions and 23 deletions.
diff --git a/R/README.md b/R/README.md
@@ -71,6 +71,9 @@ svm_predict_R(test_dataset = "../dataset/test_dataset.txt", model_file = "../dat
 
 *class_weight*:  {dict, ‘balanced’}, optional(default=None)\
     set the parameter C of class i to weight*C, for C-SVC
+
+*cv*: int, optional(default=-1)\
+    specify the number of folds in cross-validation, or -1 for no cross-validation. 
 
 *verbose*: bool(default=False)\
     enable verbose output. Note that this setting takes advantage of a per-process runtime setting; if enabled, ThunderSVM may not work properly in a multithreaded context.

diff --git a/R/svm.R b/R/svm.R
@@ -24,14 +24,14 @@ svm_train_R <-
 function(
 svm_type = 0, kernel = 2,degree = 3,gamma = 'auto',
 coef0 = 0.0, nu = 0.5, cost = 1.0, epsilon = 0.1,
-tol = 0.001, probability = FALSE, class_weight = 'None',
+tol = 0.001, probability = FALSE, class_weight = 'None', cv = '-1',
 verbose = FALSE, max_iter = -1, n_cores = -1, dataset = 'None', model_file = 'None'
 )
 {
 	res <- .C("train_R", as.character(dataset), as.integer(kernel), as.integer(svm_type),
 	as.integer(degree), as.character(gamma), as.double(coef0), as.double(nu),
 	as.double(cost), as.double(epsilon), as.double(tol), as.integer(probability),
-	as.character(class_weight), as.integer(length(class_weight)),
+	as.character(class_weight), as.integer(length(class_weight)), as.integer(cv),
 	as.integer(verbose), as.integer(max_iter), as.integer(n_cores), as.character(model_file))
 }
 

diff --git a/src/thundersvm/svm_R_interface.cpp b/src/thundersvm/svm_R_interface.cpp
@@ -134,7 +134,7 @@ extern "C" {
                 int* degree, char** gamma, double* coef0,
                  double* nu, double* cost, double* epsilon,
                  double* tol, int* probability,
-                char** class_weight, int* weight_length,
+                char** class_weight,  int* weight_length,int* n_fold,
                 int* verbose, int* max_iter, int* n_cores, char **model_file){
         int* succeed = new int[1];
         succeed[0] = 1;
@@ -239,29 +239,45 @@ extern "C" {
                 ind++;
             }
         }
-        vector<float_type> predict_y, test_y;
-        model->train(train_dataset, param_cmd);
-        model->save_to_file(model_file_path);
-        LOG(INFO) << "evaluating training score";
-        predict_y = model->predict(train_dataset.instances(), -1);
-        Metric *metric = nullptr;
-        switch (param_cmd.svm_type) {
-            case SvmParam::C_SVC:
-            case SvmParam::NU_SVC: {
-                metric = new Accuracy();
-                break;
-            }
-            case SvmParam::EPSILON_SVR:
-            case SvmParam::NU_SVR: {
-                metric = new MSE();
-                break;
+
+        int nr_fold = *n_fold;
+        vector<float_type> predict_y;
+        bool do_cross_validation = false;
+        if (nr_fold != -1) {
+            do_cross_validation = true;
+            predict_y = model->cross_validation(train_dataset, param_cmd, nr_fold);
+        } else {
+            model->train(train_dataset, param_cmd);
+            LOG(INFO) << "training finished";
+            model->save_to_file(model_file_path);
+            //   LOG(INFO) << "evaluating training score";
+            //   predict_y = model->predict(train_dataset.instances(), -1);
+        }
+//        vector<float_type> predict_y, test_y;
+//        model->train(train_dataset, param_cmd);
+//        model->save_to_file(model_file_path);
+//        LOG(INFO) << "evaluating training score";
+//        predict_y = model->predict(train_dataset.instances(), -1);
+        if(do_cross_validation) {
+            Metric *metric = nullptr;
+            switch (param_cmd.svm_type) {
+                case SvmParam::C_SVC:
+                case SvmParam::NU_SVC: {
+                    metric = new Accuracy();
+                    break;
+                }
+                case SvmParam::EPSILON_SVR:
+                case SvmParam::NU_SVR: {
+                    metric = new MSE();
+                    break;
+                }
+                case SvmParam::ONE_CLASS: {
+                }
             }
-            case SvmParam::ONE_CLASS: {
+            if (metric) {
+                std::cout << metric->name() << " = " << metric->score(predict_y, train_dataset.y()) << std::endl;
             }
         }
-        if (metric) {
-            std::cout << metric->name() << " = " << metric->score(predict_y, train_dataset.y()) << std::endl;
-        }
         return succeed;
 //        model->train(train_dataset, param_cmd);
 //        model->save_to_file(model_file_path);