def findLossAndSplit(x,y):
# 我們用 x 來表示訓練數據
# 我們用 y 來表示訓練數據的label
# x[i]表示訓練數據的第i個特征
# x_i 表示第i個訓練樣本

# minLoss 表示最小的損失 minLoss = Integet.max_value # feature 表示是訓練的數據第幾緯度的特征 feature = 0 # split 表示切分點的個數 split = 0# M 表示 樣本x的特征個數 for j in range(0,M):# 該維特征下，特征值的每個切分點，這里具體的切分方式可以自己定義for c in range(0,x[j]):L = 0# 第一類R1 = {x|x[j] <= c}# 第二類R2 = {x|x[j] > c}# 屬于第一類樣本的y值的平均值y1 = ave{y|x 屬于 R1}# 屬于第二類樣本的y值的平均值y2 = ave{y| x 屬于 R2}# 遍歷所有的樣本，找到 loss funtion 的值for x_1 in all xif x_1 屬于 R1： L += (y_1 - y1)^2 else:L += (y_1 - y2)^2if L < minLoss:minLoss = Lfeature = isplit = c return minLoss,feature ,split

定義訓練數據

train_data = [[5.1,3.5,1.4,0.2],[4.9,3.0,1.4,0.2],[7.0,3.2,4.7,1.4],[6.4,3.2,4.5,1.5],[6.3,3.3,6.0,2.5],[5.8,2.7,5.1,1.9]]

定義label

label_data = [[1,0,0],[1,0,0],[0,1,0],[0,1,0],[0,0,1],[0,0,1]]

index 表示的第幾類

def findBestLossAndSplit(train_data,label_data,index):
sample_numbers = len(label_data)
feature_numbers = len(train_data[0])
current_label = []

# define the minLossminLoss = 10000000# feature represents the dimensions of the featurefeature = 0# split represents the detail split valuesplit = 0# get current labelfor label_index in range(0,len(label_data)):current_label.append(label_data[label_index][index])# trans all featuresfor feature_index in range(0,feature_numbers):## current feature valuecurrent_value = []for sample_index in range(0,sample_numbers):current_value.append(train_data[sample_index][feature_index])L = 0## different split valueprint current_valuefor index in range(0,len(current_value)):R1 = []R2 = []y1 = 0y2 = 0for index_1 in range(0,len(current_value)):if current_value[index_1] < current_value[index]:R1.append(index_1)else:R2.append(index_1)## calculate the samples for first classsum_y = 0for index_R1 in R1:sum_y += current_label[index_R1]if len(R1) != 0:y1 = float(sum_y) / float(len(R1))else:y1 = 0## calculate the samples for second classsum_y = 0for index_R2 in R2:sum_y += current_label[index_R2]if len(R2) != 0:y2 = float(sum_y) / float(len(R2))else:y2 = 0## trans all samples to find minium loss and best splitfor index_2 in range(0,len(current_value)):if index_2 in R1:L += float((current_label[index_2]-y1))*float((current_label[index_2]-y1))else:L += float((current_label[index_2]-y2))*float((current_label[index_2]-y2))if L < minLoss:feature = feature_indexsplit = current_value[index]minLoss = Lprint "minLoss"print minLossprint "split"print splitprint "feature"print featurereturn minLoss,split,feature

findBestLossAndSplit(train_data,label_data,0)

總結

以上是生活随笔為你收集整理的机器学习算法GBDT的面试总结的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。