找到数据集上最佳的单层决策树 Parameters: dataArr - 数据矩阵 classLabels - 数据标签 D - 样本权重 Returns: bestStump - 最佳单层决策树信息 minError - 最小误差 bestClasEst - 最佳的分类结果
(dataArr,classLabels,D)
| 48 | |
| 49 | |
| 50 | def buildStump(dataArr,classLabels,D): |
| 51 | """ |
| 52 | 找到数据集上最佳的单层决策树 |
| 53 | Parameters: |
| 54 | dataArr - 数据矩阵 |
| 55 | classLabels - 数据标签 |
| 56 | D - 样本权重 |
| 57 | Returns: |
| 58 | bestStump - 最佳单层决策树信息 |
| 59 | minError - 最小误差 |
| 60 | bestClasEst - 最佳的分类结果 |
| 61 | """ |
| 62 | dataMatrix = np.mat(dataArr); labelMat = np.mat(classLabels).T |
| 63 | m,n = np.shape(dataMatrix) |
| 64 | numSteps = 10.0; bestStump = {}; bestClasEst = np.mat(np.zeros((m,1))) |
| 65 | minError = float('inf') #最小误差初始化为正无穷大 |
| 66 | for i in range(n): #遍历所有特征 |
| 67 | rangeMin = dataMatrix[:,i].min(); rangeMax = dataMatrix[:,i].max() #找到特征中最小的值和最大值 |
| 68 | stepSize = (rangeMax - rangeMin) / numSteps #计算步长 |
| 69 | for j in range(-1, int(numSteps) + 1): |
| 70 | for inequal in ['lt', 'gt']: #大于和小于的情况，均遍历。lt:less than，gt:greater than |
| 71 | threshVal = (rangeMin + float(j) * stepSize) #计算阈值 |
| 72 | predictedVals = stumpClassify(dataMatrix, i, threshVal, inequal)#计算分类结果 |
| 73 | errArr = np.mat(np.ones((m,1))) #初始化误差矩阵 |
| 74 | errArr[predictedVals == labelMat] = 0 #分类正确的,赋值为0 |
| 75 | weightedError = D.T * errArr #计算误差 |
| 76 | # print("split: dim %d, thresh %.2f, thresh ineqal: %s, the weighted error is %.3f" % (i, threshVal, inequal, weightedError)) |
| 77 | if weightedError < minError: #找到误差最小的分类方式 |
| 78 | minError = weightedError |
| 79 | bestClasEst = predictedVals.copy() |
| 80 | bestStump['dim'] = i |
| 81 | bestStump['thresh'] = threshVal |
| 82 | bestStump['ineq'] = inequal |
| 83 | return bestStump, minError, bestClasEst |
| 84 | |
| 85 | def adaBoostTrainDS(dataArr, classLabels, numIt = 40): |
| 86 | """ |
no test coverage detected