cross-validation now predict y in original order in dataset

Xtra-Computing · Nov 17, 2017 · 0906293 · 0906293
1 parent 2434965
commit 0906293
Show file tree

Hide file tree

Showing 4 changed files with 20 additions and 20 deletions.
diff --git a/include/thundersvm/util/common.h b/include/thundersvm/util/common.h
@@ -35,7 +35,7 @@ const int NUM_BLOCKS = 32 * 56;
     cudaError_t error = condition; \
     CHECK_EQ(error, cudaSuccess) << " " << cudaGetErrorString(error); \
   } while (0)
-#define SAFE_KERNEL_LAUNCH(kernel_name, ...)\
+#define SAFE_KERNEL_LAUNCH(kernel_name, ...) \
     kernel_name<<<NUM_BLOCKS,BLOCK_SIZE>>>(__VA_ARGS__);\
     CUDA_CHECK(cudaPeekAtLastError())
 #define KERNEL_LOOP(i, n) \

diff --git a/src/thundersvm/kernelmatrix.cpp b/src/thundersvm/kernelmatrix.cpp
@@ -142,10 +142,12 @@ void KernelMatrix::get_dot_product(const DataSet::node2d &instances, SyncData<fl
     for (int i = 0; i < instances.size(); ++i) {
         float_type sum = 0;
         for (int j = 0; j < instances[i].size(); ++j) {
-            CHECK_LE(instances[i][j].index, n_features_)
-                << "the number of features in testing set is larger than training set";
-            dense_ins[(instances[i][j].index - 1) * instances.size() + i] = instances[i][j].value;
-            sum += instances[i][j].value * instances[i][j].value;
+            if (instances[i][j].index < n_features_) {
+                dense_ins[(instances[i][j].index - 1) * instances.size() + i] = instances[i][j].value;
+                sum += instances[i][j].value * instances[i][j].value;
+            } else {
+//                LOG(WARNING)<<"the number of features in testing set is larger than training set";
+            }
         }
     }
     dns_csr_mul(dense_ins, instances.size(), dot_product);

diff --git a/src/thundersvm/model/svmmodel.cpp b/src/thundersvm/model/svmmodel.cpp
@@ -28,13 +28,13 @@ void SvmModel::model_setup(const DataSet &dataset, SvmParam &param) {
 vector<float_type> SvmModel::cross_validation(DataSet dataset, SvmParam param, int n_fold) {
     dataset.group_classes(this->param.svm_type == SvmParam::C_SVC);//group classes only for classification
 
-    vector<float_type> y_test_all;
-    vector<float_type> y_predict_all;
+    vector<float_type> y_predict_all(dataset.n_instances());
 
     for (int k = 0; k < n_fold; ++k) {
         LOG(INFO) << n_fold << " fold cross-validation(" << k + 1 << "/" << n_fold << ")";
         DataSet::node2d x_train, x_test;
         vector<float_type> y_train, y_test;
+        vector<int> test_idx;
         for (int i = 0; i < dataset.n_classes(); ++i) {
             int fold_test_count = dataset.count()[i] / n_fold;
             vector<int> class_idx = dataset.original_index(i);
@@ -44,6 +44,7 @@ vector<float_type> SvmModel::cross_validation(DataSet dataset, SvmParam param, i
             for (int j: vector<int>(idx_begin, idx_end)) {
                 x_test.push_back(dataset.instances()[j]);
                 y_test.push_back(dataset.y()[j]);
+                test_idx.push_back(j);
             }
             class_idx.erase(idx_begin, idx_end);
             for (int j:class_idx) {
@@ -54,12 +55,12 @@ vector<float_type> SvmModel::cross_validation(DataSet dataset, SvmParam param, i
         DataSet train_dataset(x_train, dataset.n_features(), y_train);
         this->train(train_dataset, param);
         vector<float_type> y_predict = this->predict(x_test, 1000);
-        y_test_all.insert(y_test_all.end(), y_test.begin(), y_test.end());
-        y_predict_all.insert(y_predict_all.end(), y_predict.begin(), y_predict.end());
+        CHECK_EQ(y_predict.size(), test_idx.size());
+        for (int i = 0; i < y_predict.size(); ++i) {
+            y_predict_all[test_idx[i]] = y_predict[i];
+        }
     }
-    vector<float_type> test_predict = y_test_all;
-    test_predict.insert(test_predict.end(), y_predict_all.begin(), y_predict_all.end());
-    return test_predict;
+    return y_predict_all;
 }
 
 

diff --git a/src/thundersvm/thundersvm-train.cpp b/src/thundersvm/thundersvm-train.cpp
@@ -58,17 +58,13 @@ int main(int argc, char **argv) {
     CUDA_CHECK(cudaSetDevice(parser.gpu_id));
 #endif
 
-    vector<float_type> predict_y, test_y;
+    vector<float_type> predict_y;
     if (parser.do_cross_validation) {
-        vector<float_type> test_predict = model->cross_validation(train_dataset, parser.param_cmd, parser.nr_fold);
-	   uint dataset_size = test_predict.size() / 2;
-	   test_y.insert(test_y.end(), test_predict.begin(), test_predict.begin() + dataset_size);
-	   predict_y.insert(predict_y.end(), test_predict.begin() + dataset_size, test_predict.end());
+        predict_y = model->cross_validation(train_dataset, parser.param_cmd, parser.nr_fold);
     } else {
         model->train(train_dataset, parser.param_cmd);
         model->save_to_file(parser.model_file_name);
     	predict_y = model->predict(train_dataset.instances(), 10000);
-		test_y = train_dataset.y();
     }
 
     //perform svm testing
@@ -88,8 +84,9 @@ int main(int argc, char **argv) {
         }
     }
     if (metric) {
-        LOG(INFO) << metric->name() << " = " << metric->score(predict_y, test_y); 
+        LOG(INFO) << metric->name() << " = " << metric->score(predict_y, train_dataset.y());
     }
-    return 0;
+    delete model;
+    delete metric;
 }