如何在 R 中使用 xgboost 获得三类中每一类的预测概率？

Question

我正在尝试使用 iris 数据集训练 xgboost 模型。训练代码如下所示，两个预测函数产生相同的结果。然而，结果的长度是135，而test_data只有45行。另外，结果看起来像是预测概率，但标签中有 3 个类别，而结果只产生一个向量，而不是三个类别的预测概率矩阵。那么，如何获得每个类别的预测概率以及预测类别呢？

data("iris")
iris$Species <- as.numeric(as.factor(iris$Species)) - 1

indexes <- caret::createDataPartition(iris$Species, p = .7, list = F)
train_data <- iris[indexes, ]
test_data <- iris[-indexes, ]

xgb.train <- xgb.DMatrix(data = as.matrix(train_data), label = train_data$Species)
xgb.test <- xgb.DMatrix(data = as.matrix(test_data), label = test_data$Species)

params = list("objective" = "multi:softprob", 
              "eval_metric" = "mlogloss",
              "num_class" = 3)

xgb.model <- xgboost::xgb.train(params = params, data = xgb.train, nrounds = 1000)
predict(xgb.model, newdata = xgb.test)
predict(xgb.model, newdata = xgb.test, type = "prob")

0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.985415220 0.008038994 0.006545801 0.977108896 0.016400522 0.006490625 0.985415220 0.008038994 0.006545801 0.008124468 0.983585954 0.008289632 0.005110676 0.989674747 0.005214573 0.003452316 0.993025184 0.003522499 0.005499140 0.988889933 0.005610934 0.011182932 0.977406859 0.011410273 0.005110676 0.989674747 0.005214573 0.011182932 0.977406859 0.011410273 0.011182932 0.977406859 0.011410273 0.003452316 0.993025184 0.003522499 0.010401487 0.978985548 0.010612942 0.005250969 0.005771303 0.988977730 0.005250969 0.005771303 0.988977730 0.005250969 0.005771303 0.988977730 0.005239322 0.007976402 0.986784279 0.005239322 0.007976402 0.986784279 0.005239322 0.007976402 0.986784279 0.005250969 0.005771303 0.988977730 0.005219116 0.011802264 0.982978642 0.005250969 0.005771303 0.988977730 0.005219116 0.011802264 0.982978642 0.005219116 0.011802264 0.982978642 0.005250969 0.005771303 0.988977730 0.005250969 0.005771303 0.988977730 0.005250969 0.005771303 0.988977730 0.005180326 0.019146746 0.975672841

Answer 1

预测值连接成一个向量。您可以简单地将它们转换为矩阵以获得每个类别的预测值。

pred <- predict(xgb.model, newdata = xgb.test)

pred <- matrix(pred, ncol=xgb.model$params$num_class, byrow=TRUE)
head(pred)

          [,1]        [,2]        [,3]
[1,] 0.9858927 0.007713272 0.006394033
[2,] 0.9858927 0.007713272 0.006394033
[3,] 0.9858927 0.007713272 0.006394033
[4,] 0.9858927 0.007713272 0.006394033
[5,] 0.9858927 0.007713272 0.006394033
[6,] 0.9790474 0.014603053 0.006349638