knn代码研究 opencv3 #

knn还不懂。结果不能解释。

代码如下：

float[] trainFeaturesData =
            {
                 2,2,2,2,
                 3,3,3,3,
                 4,4,4,4,
                 5,5,5,5,
                 6,6,6,6,
                 7,7,7,7
            };
            var trainFeatures = new Mat(6, 4, MatType.CV_32F, trainFeaturesData);

            int[] trainLabelsData = { 2, 3, 4, 5, 6, 7 };
            var trainLabels = new Mat(1, 6, MatType.CV_32F, trainLabelsData);

            var kNearest = KNearest.Create();
            kNearest.Train(trainFeatures, SampleTypes.RowSample, trainLabels);

            float[] testFeatureData = { 3, 3, 3, 3 };
            var testFeature = new Mat(1, 4, MatType.CV_32F, testFeatureData);

            const int k = 1;
            var results = new Mat();
            var neighborResponses = new Mat();
            var dists = new Mat();
            var detectedClass = (int)kNearest.FindNearest(testFeature, k, results, neighborResponses, dists);

            /*
            K=1
            [3]
            [0]

            K=4:
            [3, 2, 4, 5]
            [0, 4, 4, 16]
            */
            Console.WriteLine("results: {0},{1},{2},{3}", results.Get<int>(0,0), results.Get<int>(0, 1), results.Get<int>(0, 2), results.Get<int>(0, 3));
            //Console.WriteLine("DetectedClass: {0}", detectedClass);
            Console.WriteLine("neighborResponses: {0}", neighborResponses);
            Console.WriteLine("dists: {0}", dists);

这个代码思路流畅，但结果充满了怪异

程序首先构建了一个训练数据集合trainFeatures ，然后构建了一个标签。然后训练。接着构建待测试矩阵，然后用knn找邻居，

K等于1时results: 3,0,450397088,0

2： results: 2,1,0,0

3：2,1,-2095420776,1

4：2,0,458219152,0

这里的结果让人困惑。同样奇葩的是 int res = (int)kNearest.Predict(testFeature, results);的res始终是0，不知道哪里能得到解释。