Kotlin OpenCV 机器学习70 DTrees 梯度提升树

1 OpenCV 机器学习算法

2 OpenCV 深度学习算法

3 OpenCV 无监督学习算法

4 Kotlin 梯度提升树

Kotlin OpenCV 代码

package com.xu.com.xu.ml

import cn.hutool.core.util.CharsetUtil
import cn.hutool.extra.compress.CompressUtil
import org.opencv.core.Core
import org.opencv.core.CvType
import org.opencv.core.Mat
import org.opencv.core.Size
import org.opencv.imgcodecs.Imgcodecs
import org.opencv.imgproc.Imgproc
import org.opencv.ml.DTrees
import org.opencv.ml.Ml
import java.io.File
import java.util.*


object Train {

    init {
        val os = System.getProperty("os.name")
        val type = System.getProperty("sun.arch.data.model")
        if (os.uppercase(Locale.getDefault()).contains("WINDOWS")) {
            val lib = if (type.endsWith("64")) {
                File("lib\\opencv-4.9\\x64\\" + System.mapLibraryName(Core.NATIVE_LIBRARY_NAME))
            } else {
                File("lib\\opencv-4.9\\x86\\" + System.mapLibraryName(Core.NATIVE_LIBRARY_NAME))
            }
            System.load(lib.absolutePath)
        }
        println(Core.VERSION)
    }

    @JvmStatic
    fun main(args: Array<String>) {
        val (trainImages, trainLabels) = load("lib/data/image/train/")
        val (testImages, testLabels) = load("lib/data/image/predict/")

        // 梯度提升树
        val model = DTrees.create()
        model.maxDepth = 20
        model.minSampleCount = 2
        model.useSurrogates = false
        model.cvFolds = 0
        model.use1SERule = false
        model.truncatePrunedTree = false
        model.regressionAccuracy = 0.01f

        // 转换为OpenCV的Mat格式
        val trainImagesData = Mat(trainImages.size, 784, CvType.CV_32F)
        trainImages.forEachIndexed { index, floatArray ->
            trainImagesData.put(index, 0, floatArray)
        }
        val trainLabelsData = Mat(trainLabels.size, 1, CvType.CV_32S)
        trainLabelsData.put(0, 0, trainLabels.toIntArray())

        // 训练模型
        model.train(trainImagesData, Ml.ROW_SAMPLE, trainLabelsData)
        model.save("lib/data/image/ml/DTrees.xml")

        // 评估训练集准确率
        val train = accuracy(model, trainImages, trainLabels)
        println("训练集准确率: $train")

        // 评估测试集准确率
        val test = accuracy(model, testImages, testLabels)
        println("测试集准确率: $test")
    }

    /**
     * 加载数据
     */
    private fun load(path: String): Pair<List<FloatArray>, List<Int>> {
        val images = mutableListOf<FloatArray>()
        val labels = mutableListOf<Int>()

        for (i in 0..9) {
            val dir = File("$path/$i")
            dir.listFiles()?.forEach { file ->
                val img = Imgcodecs.imread(file.absolutePath, Imgcodecs.IMREAD_GRAYSCALE)
                if (!img.empty()) {
                    Imgproc.resize(img, img, Size(28.0, 28.0))
                    val array = ByteArray(784)
                    img.get(0, 0, array)
                    images.add(array.map { it / 255.0f }.toFloatArray())
                    labels.add(i)
                }
            }
        }
        return Pair(images, labels)
    }

    /**
     * 计算准确率
     */
    private fun accuracy(model: DTrees, images: List<FloatArray>, labels: List<Int>): Double {
        var correct = 0
        images.forEachIndexed { index, image ->
            val sample = Mat(1, 784, CvType.CV_32F)
            sample.put(0, 0, image)
            val response = model.predict(sample)
            if (response.toInt() == labels[index]) {
                correct++
            }
        }
        return correct.toDouble() / images.size * 100
    }

    private fun unzip() {
        // 解压训练图片
        CompressUtil.createExtractor(
            CharsetUtil.defaultCharset(),
            File("lib/data/image/train.7z")
        ).extract(File("lib/data/image/train/"))
        // 解压测试图片
        CompressUtil.createExtractor(
            CharsetUtil.defaultCharset(),
            File("lib/data/image/predict.7z")
        ).extract(File("lib/data/image/predict/"))
    }

}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121

5 Kotlin 梯度提升树 训练结果

4.9.0
训练集准确率: 98.19333333333333
测试集准确率: 84.54
1
2
3