使用 R 中的一个类的最小值来估算缺失值答案

【问题标题】：impute missing values using minimum of a class in R使用 R 中的一个类的最小值来估算缺失值
【发布时间】：2015-11-08 06:40:10
【问题描述】：

我是 R 新手，需要帮助我在我目前正在处理的数据集中的一列中估算缺失值。下图显示了我想要估算的缺失值以及几列。

我想使用客户以前的条目为客户填写最小数量的值，因为我认为这最适合我的情况和数据。例如，在图像中，我应该能够用 1 填充缺失值（最小值为 1、5、2）。

在我的搜索过程中，我主要遇到了对给定类使用均值的方法，而不是最小值或最大值。

任何帮助或指点将不胜感激。

编辑：这是 dput 的输出。

structure(list(YEAR = c(2011L, 2012L, 2014L, 2015L, 2011L, 2012L
), CustomerId = c("00000063", "00000063", "00000063", "00000063", 
"00000065", "00000065"), MemberType = structure(c(2L, 2L, 2L, 
2L, 2L, 2L), .Label = c("GROUP", "INDIVIDUAL", "PARTNER"), class = "factor"), 
    MembershipTypeCode = structure(c(6L, 6L, 6L, 10L, 6L, 6L), .Label = c("EGROUP", 
    "EINDIV", "EINDIV2", "EPARTNER", "GROUP", "INDIV", "INDIV2", 
    "INDIV3", "PARTNER", "PLUS", "PLUS2", "PLUS20", "PLUS3", 
    "PLUSENTERPRI", "PLUSGROUP", "PLUSGROUP2", "PROF_ENTERPR", 
    "PROF_GROUP", "PROF_GROUP2", "PROF_INDIV", "PROF_INDIV2", 
    "PROF_INDIV3"), class = "factor"), MembershipPeriodBegin = structure(c(15279, 
    15677, 16071, 16436, 15006, 15371), class = "Date"), MembershipPeriodEnd = structure(c(15644, 
    16070, 16435, 16800, 15370, 15736), class = "Date"), ConsecutiveYearsAsMember = c(14L, 
    15L, 17L, 18L, 8L, 9L), AllocationUsage = c(0, 0, 0, 0, 0, 
    0), SetCOPPreference = structure(c(2L, 2L, 2L, 2L, 2L, 2L
    ), .Label = c("Y", "N"), class = "factor"), Purchase.Qty = c(2L, 
    5L, 1L, NA, 7L, 27L), Webcast.Registration = c(0L, 0L, 0L, 
    0L, 0L, 1L), Web.Visits = c(0L, 0L, 42L, 0L, 0L, 0L), Web.Page.Views = c(0L, 
    0L, 98L, 0L, 0L, 0L), Blog.Visits = c(0L, 0L, 3L, 0L, 0L, 
    0L), Blog.Page.Views = c(0L, 0L, 4L, 0L, 0L, 0L), Forum.Visits = c(0L, 
    0L, 45L, 0L, 0L, 0L), Forum.Page.Views = c(0L, 0L, 102L, 
    0L, 0L, 0L), ParatureTickets = c(0L, 0L, 0L, 0L, 0L, 0L), 
    ParatureChats = c(0L, 0L, 0L, 0L, 0L, 0L), Registered.for.Edu = c(0L, 
    0L, 0L, 0L, 0L, 0L), Attended.ICE = structure(c(2L, 2L, 2L, 
    2L, 2L, 2L), .Label = c("Y", "N"), class = "factor"), Attended.TK = structure(c(2L, 
    2L, 2L, 2L, 2L, 2L), .Label = c("Y", "N"), class = "factor"), 
    Frugal = structure(c(2L, 2L, 2L, 2L, 2L, 2L), .Label = c("Y", 
    "N"), class = "factor"), Chapter.Board = structure(c(2L, 
    2L, 2L, 2L, 2L, 2L), .Label = c("Y", "N"), class = "factor"), 
    Retained = structure(c(5L, 5L, 5L, 1L, 5L, 5L), .Label = c("Active", 
    "Awaiting Renewal", "Future Dated", "Lost", "Retained"), class = "factor"), 
    ProfileCompletion = c(60, 60, 60, 60, 60, 60), NumberofLogins = c(1L, 
    1L, 15L, 0L, 0L, 4L), Downloads = c(NA_integer_, NA_integer_, 
    NA_integer_, NA_integer_, NA_integer_, NA_integer_), ForumMember = structure(c(NA_integer_, 
    NA_integer_, NA_integer_, NA_integer_, NA_integer_, NA_integer_
    ), .Label = "N", class = "factor"), FreeUpgrade = structure(c(1L, 
    1L, 1L, 1L, 1L, 1L), .Label = c("Y", "N"), class = "factor")), .Names = c("YEAR", 
"CustomerId", "MemberType", "MembershipTypeCode", "MembershipPeriodBegin", 
"MembershipPeriodEnd", "ConsecutiveYearsAsMember", "AllocationUsage", 
"SetCOPPreference", "Purchase.Qty", "Webcast.Registration", "Web.Visits", 
"Web.Page.Views", "Blog.Visits", "Blog.Page.Views", "Forum.Visits", 
"Forum.Page.Views", "ParatureTickets", "ParatureChats", "Registered.for.Edu", 
"Attended.ICE", "Attended.TK", "Frugal", "Chapter.Board", "Retained", 
"ProfileCompletion", "NumberofLogins", "Downloads", "ForumMember", 
"FreeUpgrade"), row.names = c(NA, 6L), class = "data.frame")

谢谢，
普拉提克

【问题讨论】：

请编辑问题以插入dput(head(data)) 的输出，以便任何人都可以使用数据集。我怀疑data$Purchase.Qty<-ave(data$Purchase.Qty,data$CustomerId,FUN=function(x) {x[is.na(x)]<-min(x,na.rm=TRUE);x}) 可能有效，但如果没有可重现的示例就无法测试。
它给了我以下警告信息'In min(x, na.rm = TRUE) : no non-missing arguments to min;返回 Inf'

标签： r missing-data

【解决方案1】：

我们可以使用na.aggregate 和FUN= min。我们将“data.frame”转换为“data.table”（setDT(df1)），按“CustomerID”分组，我们在“PurchaseQty”上应用na.aggregate，并将（:=）输出分配回“PurchaseQty” .

library(data.table)
library(zoo)
setDT(df1)[, PurchaseQty := na.aggregate(PurchaseQty, FUN= min) , by = CustomerID]

数据

df1 <- data.frame(CustomerID= rep(1:2, each=4), PurchaseQty= c(4, 3, NA, 3, 1, 9, NA, 4))

【讨论】：

【解决方案2】：

由于您没有提供数据，这里是一个玩具示例，我将如何在基础 R 中做到这一点：

# simple sample data
data <- data.frame( a = rep( 10:12, each = 4 ), b = 12:1 )
data[ c( 3, 5, 12 ), 2 ] <- NA

# for each unique a value, get the row index with the min b value, 
# and write that min value to col b where b is NA
for( i in unique( data$a ) )
    data[ which( is.na( data$b ) & data$a == i  ), "b" ] <-
        min( data[ data$a == i, "b" ], na.rm = TRUE )

data
    a  b
1  10 12
2  10 11
3  10  9
4  10  9
5  11  5
6  11  7
7  11  6
8  11  5
9  12  4
10 12  3
11 12  2
12 12  2

【讨论】：

虽然这行得通...我只是想知道没有任何估算包可以做到这一点吗？
我不知道（但我知道的很少......） - 也许@akrun 的建议接近这个预期。但是您可以将我的循环变成一个自定义函数，将 data.frame 和列用作参数并返回 data.frame 的“估算版本”。