Question

对于一个回归项目，我使用kmeans通过nbclust创建了数据集群。该项目目前有4个集群，但如果成功，该项目可能涉及分析多个美国城市。尽管我可以手动构建每个回归，但我希望减少代码量，从而减少编码错误的机会。

当前模型使用具有多个子集和10-15个不同自变量的glm。我目前正在手动建立每个聚类回归。

Tulsa$Cluster <- Tulsa$Best.partition
#This creates a vector of clusters; there are currently 4 distinct clusters

summary(Tulsa_Cluster1 <- glm(formula = Tulsa$Result ~ Tulsa$FactorA Tulsa$FactorB + Tulsa$FactorC + Tulsa$FactorD + Tulsa$FactorE + Tulsa$FactorF, 
                              method = "glm.fit", family = gaussian(),subset = Tulsa$Cluster =="1"))

summary(Tulsa_Step <- stepAIC(Tulsa_Cluster1, direction = "both",trace = FALSE, 
                              scope=list(lower = ~ Tulsa_Cluster1$FactorA)))

我希望通过聚类进行回归，然后可以对其进行总结。

Answer 1

请考虑使用unique，levels，split和by进行以下任意一种方法来迭代 Cluster 的唯一值：< / p>

unique + for

for (c in unique(Tulsa$Cluster)) {
   Tulsa_Cluster <- glm(formula = Result ~ FactorA + FactorB + FactorC + FactorD + FactorE + FactorF, 
                        data = Tulsa, method = "glm.fit", family = gaussian(), subset = Cluster == c)
   print(summary(Tulsa_Cluster))  # REQUIRED IN for LOOPS TO FOR CONSOLE OUTPUT

   Tulsa_Step <- stepAIC(Tulsa_Cluster, direction = "both", trace = FALSE, 
                         scope=list(lower = ~ Tulsa_Cluster$FactorA))
   print(summary(Tulsa_Step))     # REQUIRED IN for LOOPS TO FOR CONSOLE OUTPUT
}

levels + for

for (c in levels(factor(Tulsa$Cluster))) {
   Tulsa_Cluster <- glm(formula = Result ~ FactorA + FactorB + FactorC + FactorD + FactorE + FactorF, 
                        data = Tulsa, method = "glm.fit", family = gaussian(), subset = Cluster == c)
   print(summary(Tulsa_Cluster))

   Tulsa_Step <- stepAIC(Tulsa_Cluster, direction = "both", trace = FALSE, 
                         scope=list(lower = ~ Tulsa_Cluster$FactorA))
   print(summary(Tulsa_Step))
}

下面两个返回命名对象列表。

split + lapply （无子集arg）

data_list <- split(Tulsa, Tulsa$Cluster)

model_list <- lapply(data_list, function(sub_df) {
   Tulsa_Cluster <- glm(formula = Result ~ FactorA + FactorB + FactorC + FactorD + FactorE + FactorF, 
                        data = sub_df, method = "glm.fit", family = gaussian())

   Tulsa_Step <- stepAIC(Tulsa_Cluster, direction = "both", trace = FALSE, 
                         scope=list(lower = ~ Tulsa_Cluster$FactorA))

   # RETURN BOTH SUMMARIES INTO LIST
   return(list(Tulsa_Cluster, Tulsa_Step))
})

by （无子集arg）

model_list <- by(Tulsa, Tulsa$Cluster, function(sub_df) {
   Tulsa_Cluster <- glm(formula = Result ~ FactorA + FactorB + FactorC + FactorD + FactorE + FactorF, 
                        data = sub_df, method = "glm.fit", family = gaussian())

   Tulsa_Step <- stepAIC(Tulsa_Cluster, direction = "both", trace = FALSE, 
                         scope=list(lower = ~ Tulsa_Cluster$FactorA))

   # RETURN BOTH SUMMARIES INTO LIST
   return(list(Tulsa_Cluster, Tulsa_Step))
})

有没有一种方法可以按子集循环进行回归？

1 个答案: