{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# 机器学习与社会科学应用" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 第六章 无监督学习算法" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 第二节 K-means聚类" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "郭峰 \n", " 教授、博士生导师 \n", "上海财经大学公共经济与管理学院 \n", "上海财经大学数实融合与智能治理实验室 \n", " 邮箱：guofengsfi@163.com " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "本节目录 \n", "2.1.演示性案例 \n", "2.2.手写数字识别 \n", "2.3.K-meas参数 " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 2.1. 演示性案例" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "import matplotlib.pyplot as plt\n", "from sklearn.datasets import load_iris\n", "from sklearn.cluster import KMeans\n", "\n", "# 加载鸢尾花数据集\n", "iris = load_iris()\n", "X = iris.data # 特征矩阵\n", "X[0:5]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 创建K-means模型\n", "kmeans = KMeans(n_clusters=3, random_state=0)\n", "\n", "# 拟合模型到数据\n", "kmeans.fit(X)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 获取聚类中心点的坐标\n", "cluster_centers = kmeans.cluster_centers_\n", "print(\"Cluster centers:\")\n", "print(cluster_centers)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 预测每个数据点的簇标签\n", "labels = kmeans.labels_\n", "print(\"Labels:\")\n", "print(labels)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 可视化聚类结果\n", "plt.scatter(X[:, 0], X[:, 1], c=labels)\n", "plt.scatter(cluster_centers[:, 0], cluster_centers[:, 1], marker='x', s=200, linewidths=3, color='r')\n", "plt.xlabel(\"Feature 1\")\n", "plt.ylabel(\"Feature 2\")\n", "plt.title(\"K-means Clustering\")\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 2.2 手写数字识别" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from numpy import *\n", "import os\n", "from os import listdir\n", "import operator\n", "from sklearn.cluster import KMeans\n", "from sklearn import metrics\n", "\n", "#步骤1：路径与标签\n", "path1='D:/python/机器学习与社会科学应用/演示数据/06无监督学习/digits/trainingDigits/'\n", "path2='D:/python/机器学习与社会科学应用/演示数据/06无监督学习/digits/testDigits/'\n", "print(path1)\n", "#将训练数据存储到一个矩阵中1024维，并存储对应的标签\n", "trainName=listdir(path1)\n", "trainNum=len(trainName)\n", "trainNumpy = zeros((trainNum,1024))\n", "#print(\"trainNum=%d\"%trainNum)\n", "#对文件名进行分析，训练文本对应的标签\n", "print(trainNum)\n", "print(trainName[0])\n", "\n", "\n", "#将测试数据存储到一个矩阵中1024维，并存储对应的标签\n", "testName=listdir(path2)\n", "testNum=len(testName)\n", "testNumpy = zeros((testNum,1024))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "#步骤2：将像素数据转换成向量数据\n", "#这个函数是1024长度的向量\n", "def img2vector(filename):\n", " vect=zeros((1,1024))\n", " f=open(filename)\n", " for i in range(32):\n", " line=f.readline()\n", " for j in range(32):\n", " vect[0,32*i+j]=int(line[j])\n", " return vect\n", "handlabel=[]\n", "\n", "#训练集\n", "for i in range(trainNum):\n", " filename=trainName[i]#文件名\n", " filestr=filename.split('.')[0]#不带后缀的文件名\n", " filelabel=int(filestr.split('_')[0])#文件的标签,0,1,2,....,9\n", " #将标签添加至handlabel中\n", " handlabel.append(filelabel)\n", " trainNumpy[i,:]=img2vector(path1+str(filename))#转成1024\n", " #print(handlabel[:20])\n", "\n", "print(trainNumpy[0][0:40])\n", "print(trainNumpy.shape[0])\n", "print(handlabel[0:40])\n", "print(len(handlabel))\n", "\n", "\n", "handlabel_test=[]\n", "for i in range(testNum):\n", " filename_test=testName[i]#文件名\n", " filestr_test=filename_test.split('.')[0]#不带后缀的文件名\n", " filelabel_test=int(filestr_test.split('_')[0])#文件的标签,0,1,2,....,9\n", " #将标签添加至handlabel中\n", " handlabel_test.append(filelabel_test)\n", " testNumpy[i,:]=img2vector(path1+str(filename_test))#转成1024" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "y_train=handlabel\n", "x_train=trainNumpy #列表，1934个元素，每个元素又是一个1024长度的列表\n", "print(len(x_train))\n", "print(len(y_train))\n", "kmeans = KMeans(n_clusters=10, random_state=100)\n", "kmeans.fit(x_train)\n", "#测试集\n", "print(y_train[0:189])\n", "y_pred = kmeans.predict(x_train)\n", "print(y_pred[0:189])\n", "print(1-len([y for y in y_pred[0:189] if y!=8])/len(y_train[0:189]))\n", "#print(adjusted_rand_score(y_train, y_pred))\n", "#随机试验中，把0可能会随机分配给某个\"标签\"\n", "\n", "#结果评价：根据已知分类标签\n", "for num in range(0,10):\n", " x_train_temp=[x_train[j] for j in [i for i,y in enumerate(y_train) if y==num]]\n", " y_train_temp=[y for y in y_train if y==num]\n", " y_pred_temp=kmeans.predict(x_train_temp)\n", " y_pred_dict=dict((a,list(y_pred_temp).count(a)) for a in y_pred_temp)\n", " y_pred_value=[k for k,v in y_pred_dict.items() if max(y_pred_dict.values())==v]\n", " print(\"数值%d对应的预测\\\"聚类代码\\\"为:%d,准确率为：%.4f\"\n", " %(num,y_pred_value[0],sum(y_pred_temp==y_pred_value[0])/len(y_train_temp)))\n", "#3和9没有区分开\n", "#1和8没有区分开" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "kmeans.get_params()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "#结果评价：兰德指数\n", "#https://blog.csdn.net/sinat_26917383/article/details/70577710" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "scrolled": true }, "outputs": [], "source": [ "for k in range(2,15):\n", " kmeans = KMeans(n_clusters=k)\n", " kmeans.fit(x_train)\n", " y_pred = kmeans.predict(x_train)\n", " print(k,metrics.adjusted_rand_score(y_train, y_pred))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "#轮廓系数\n", "for k in range(2,15):\n", " kmeans = KMeans(n_clusters=k)\n", " kmeans.fit(x_train)\n", " y_pred = kmeans.predict(x_train)\n", " print(k,metrics.silhouette_score(x_train, y_pred, metric='euclidean'))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 2.3. K-Means参数" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# n_clusters：整型，缺省值=8 ，生成的聚类数。\n", "# max_iter：整型，缺省值=300 。执行一次k-means算法所进行的最大迭代数。 \n", "# n_init：整型，缺省值=10 。用不同的聚类中心初始化值运行算法的次数，最终解是在inertia意义下选出的最优结果。 \n", "# init：有三个可选值：’k-means++’， ‘random’，或者传递一个ndarray向量。 此参数指定初始化方法，默认值为 ‘k-means++’。 \n", "# 　　（１）‘k-means++’ 用一种特殊的方法选定初始聚类中发，可加速迭代过程的收敛。\n", "# 　　（２）‘random’ 随机从训练数据中选取初始质心。 \n", "# 　　（３）如果传递的是一个ndarray，则应该形如 (n_clusters, n_features) 并给出初始质心。 \n", "# precompute_distances：三个可选值，‘auto’，True 或者 False。 预计算距离，计算速度更快但占用更多内存。 \n", "# 　　（１）‘auto’：如果 样本数乘以聚类数大于 12million 的话则不预计算距离。\n", "# 　　（２）True：总是预先计算距离。 \n", "# 　　（３）False：永远不预先计算距离。 \n", "# tol：float类型，默认值= 1e-4　与inertia结合来确定收敛条件。 \n", "# n_jobs：整形数。　指定计算所用的进程数。内部原理是同时进行n_init指定次数的计算。 \n", "# 　　（１）若值为 -1，则用所有的CPU进行运算。若值为1，则不进行并行运算。\n", "# 　　（２）若值小于-1，则用到的CPU数为(n_cpus + 1 + n_jobs)。因此如果 n_jobs值为-2，则用到的CPU数为总CPU数减1。 \n", "# random_state：整型或 numpy.RandomState 类型，可选 用于初始化质心的生成器（generator）。如果值为一个整数，则确定一个seed。此参数默认值为numpy的随机数生成器。 \n", "# copy_x：布尔型，默认值=True 　　\n", "# 当我们precomputing distances时，将数据中心化会得到更准确的结果。如果把此参数值设为True，则原始数据不会被改变。如果是False，则会直接在原始数据 \n", "# 上做修改并在函数返回值时将其还原。但是在计算过程中由于有对数据均值的加减运算，所以数据返回后，原始数据和计算前可能会有细小差别。" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "本节结束 " ] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.9.12" } }, "nbformat": 4, "nbformat_minor": 4 }