异常检测-基于孤立森林算法Isolation-based Anomaly Detection-3-例子
参考：https://scikit-learn.org/stable/auto_examples/ensemble/plot_isolation_forest.html#sphx-glr-auto-examples-ensemble-plot-isolation-forest-py
代码：
print(__doc__) import numpy as np import matplotlib.pyplot as plt from sklearn.ensemble import IsolationForest rng = np.random.RandomState(42) # 构建训练数据，即100个属性值为2的样本，属性的值为 随机[0,1]数*0.3 X = 0.3 * rng.randn(100, 2) # 将上面得到的值+2和-2各生成100个值在2和-2附近的样本 #拼接后训练数据大小为(200, 2) X_train = np.r_[X + 2, X - 2] #按列连接矩阵，要求列相等，行拼接 # 产生一些有规律的新观察值 X = 0.3 * rng.randn(20, 2) #拼接后训练数据大小为(40, 2) X_test = np.r_[X + 2, X - 2] # 均匀分布生成异常数据集，大小为(20, 2)，值的范围为[-4,4] X_outliers = rng.uniform(low=-4, high=4, size=(20, 2)) # 构建森林，进行采样，子采样大小为100 # 默认参数max_features=1，则每棵树都仅使用一个属性来进行切割 # 如果你想要选择多个属性（当你的数据是多维，即有多个属性时）则记得设置该参数 clf = IsolationForest(behaviour='new', max_samples=100, random_state=rng, contamination='auto') # 训练森林，选择属性和分割值等 clf.fit(X_train) #然后使用该构建好的森林进行预测 y_pred_train = clf.predict(X_train) print(y_pred_train) y_pred_test = clf.predict(X_test) print(y_pred_test) y_pred_outliers = clf.predict(X_outliers) print(y_pred_outliers) # 画图, the samples, and the nearest vectors to the plane # xx和yy大小分别为(50,50) xx, yy = np.meshgrid(np.linspace(-5, 5, 50), np.linspace(-5, 5, 50)) # 先拉直xx和yy为大小为(2500,)的一维向量 # 然后按行拼接xx,yy，即行数相等，列数增加；即两者拼成(2500,2)的坐标点 # 然后得到这几个点的异常分数 # 正常点的异常分数为整数，异常的为负数 Z = clf.decision_function(np.c_[xx.ravel(), yy.ravel()]) Z = Z.reshape(xx.shape) plt.title("IsolationForest") # 绘制网格点的异常分数的等高线图，看图可知，颜色越浅越可能为正常点，越深越为异常点 plt.contourf(xx, yy, Z, cmap=plt.cm.Blues_r) # 在等高线中标出训练点、测试点、异常点的位置，看它们是不是在对应的颜色位置 # 可见训练点和测试点都在颜色前的区域，异常点都在颜色深的区域 b1 = plt.scatter(X_train[:, 0], X_train[:, 1], c='white', s=20, edgecolor='k') b2 = plt.scatter(X_test[:, 0], X_test[:, 1], c='green', s=20, edgecolor='k') c = plt.scatter(X_outliers[:, 0], X_outliers[:, 1], c='red', s=20, edgecolor='k') plt.axis('tight') # 绘制图的坐标和图例信息 plt.xlim((-5, 5)) plt.ylim((-5, 5)) plt.legend([b1, b2, c], ["training observations", "new regular observations", "new abnormal observations"], loc="upper left") plt.show()
返回：
Automatically created module for IPython interactive environment [ 1 -1 1 -1 1 1 -1 -1 1 -1 -1 1 1 1 1 -1 1 -1 -1 1 1 1 -1 1 -1 1 1 -1 1 1 1 -1 -1 1 1 -1 1 -1 1 -1 1 -1 1 1 1 1 1 -1 1 1 1 1 1 -1 1 -1 -1 1 1 -1 1 -1 -1 1 1 -1 1 -1 1 -1 1 -1 1 -1 1 1 1 1 -1 1 1 -1 -1 -1 1 1 1 1 1 -1 1 1 1 1 -1 1 1 1 1 1 1 -1 1 -1 1 1 -1 -1 1 -1 -1 -1 1 1 1 -1 1 -1 -1 -1 1 1 -1 1 -1 1 1 -1 1 1 1 -1 -1 1 1 -1 -1 -1 1 -1 1 -1 1 1 1 1 1 -1 1 1 -1 1 1 -1 1 -1 -1 1 1 -1 1 -1 -1 -1 1 -1 1 -1 1 -1 1 -1 1 -1 1 1 1 1 -1 -1 1 -1 -1 -1 1 -1 1 1 -1 -1 1 1 1 1 -1 1 1 1 1 1] [ 1 -1 -1 1 -1 -1 -1 1 1 1 -1 -1 1 1 1 1 1 -1 -1 1 1 -1 -1 1 -1 1 -1 1 1 1 -1 -1 1 1 1 1 1 -1 -1 1] [-1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1]
图为：
如果将contamination设置为0.，表示训练数据中没有异常数据，返回为：
Automatically created module for IPython interactive environment [[-0.48822863 -3.37234895] [-3.79719405 3.70118732] [ 2.68784096 1.56779365] [-0.72837644 -2.61364544] [-2.74850366 -1.99805681] [ 0.39381332 1.71676738] [ 1.28157901 -1.76052882] [ 3.63892225 1.90317533] [ 0.43483242 0.89376597] [-0.6431995 -2.01815208] [-1.15221857 2.06276888] [-3.88485209 -3.07141888] [-3.63197886 -3.67416958] [ 2.84368467 1.62926288] [-0.20660937 -3.21732671] [-0.067073 -0.21222583] [-2.61438504 -0.52918681] [-0.81196212 0.92680078] [ 1.08074921 -3.63756792] [-1.00309908 1.00687933]] [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1] [ 1 1 1 1 1 1 1 1 1 1 -1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 -1 1 1 1 1 1 1 1 1 1] [-1 -1 -1 -1 1 -1 1 1 -1 1 -1 -1 -1 1 -1 -1 -1 -1 -1 -1]
可见其会将一些接近训练点的数据也预测为正常数据
如果同时设置构建树时使用的属性为2，即max_features=2，而不是默认的1，结果为：
[1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1] [ 1 1 1 1 1 1 1 1 1 1 -1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 -1 1 1 1 1 1 1 1 1 1] [-1 -1 -1 -1 1 -1 1 1 -1 1 -1 -1 -1 1 -1 -1 -1 -1 -1 -1]
感觉训练效果更好了，测试数据基本上都能验证为正常点
所以根据你自己的需要来配置参数吧
