访问随机林sparkR中的概率对象列

在斯巴克，我有这样的东西

rf <- spark.randomForest(train, formula, type = "classification")
pred <- predict(rf,test) 

输出是您可以在图像中看到的内容

---

如何获得概率值？

您必须使用函数

sparkR.callJMethod

为每个对象调用名为

values

的Java方法

t(sapply(collect(select(pred, "probability"))$probability, 
         FUN = function(x) sparkR.callJMethod(x, "values")))

下面是使用Iris数据集的完整示例。目标值为

物种

，有3个级别，共有150个数据点

df <- createDataFrame(iris)
model <- spark.randomForest(df, Species ~ ., type = "classification")
summary(model)

predictions <- predict(model, df)

local_prob <- collect(select(predictions, "probability"))$probability

t(sapply(local_prob, FUN = function(x) sparkR.callJMethod(x, "values")))

df <- createDataFrame(iris)
model <- spark.randomForest(df, Species ~ ., type = "classification")
summary(model)

predictions <- predict(model, df)

local_prob <- collect(select(predictions, "probability"))$probability

t(sapply(local_prob, FUN = function(x) sparkR.callJMethod(x, "values")))

       [,1]        [,2]        [,3]
  [1,] 0           0           1   
  [2,] 0           0           1   
  [3,] 0           0           1   
...  
[148,] 0           1           0   
[149,] 0.05        0.95        0   
[150,] 0.01805556  0.9819444   0