R进行数据分析

1. 排序

• sort(x, decreasing = ): 返回排序好的数据
• order(x, decreasing = ): 返回排序好的数据的索引

例子：

v = c(2, 9, 1, 45, -3, 19, -5, 6)

sort(v) # returns ordered v in decreasing order
结果：
# [1] -5 -3 1 2 6 9 19 45
sort(v, decreasing = FALSE) # orders v in increasing order
结果：
# [1] -5 -3 1 2 6 9 19 45

order(v) # returns order of the indexes in v
结果：
# [1] 7 5 3 1 8 2 6 4

order()可以和更加复杂的数据结构配合使用，从而适应更加复杂的场景
如：

#Imagine that you just want to access Sepal Length and Species.
# You can access those values in different ways:
ir[order(ir$Sepal.Length, decreasing = TRUE),c("Sepal.Length", "Species")][1:5,]

ir[order(ir$Sepal.Length, decreasing = TRUE),][1:5, c("Sepal.Length", "Species")]

2. aggregate()

2.1 分组处理数据

aggregate(X, by, FUN, . . . ,simplify = TRUE)

• X is an R object (commonly a data frame)
• by is a list of the elements by which you will be grouping your data.
• FUN is the function that will be applied to each subset.
• simplify is a logical value that indicates if results should be simplified into a vector or a matrix.

举例:

针对单一对象，一种分组
aggregate(ir$Sepal.Length, by= list(ir$Species), FUN=mean)
aggregate(ir$Sepal.Width, by= list(ir$Species), summary)

针对多个对象，一种分组
aggregate(ir[,c("Sepal.Length", "Sepal.Width")], by=list(ir$Species), mean)

针对单一对象，多种分组
mean_by_sp_ind = aggregate(ir$Sepal.Length, by=list(ir$Species, ir$indoor), mean)

2.2 另一版本使用

aggregate(formula, data, FUN, subset)

• formula details the in the manner y ~ x or cbind(y1, y2) ~ x1+x2,where y and cbind(y1,y2) are the numeric data to be split and x or x1 + x2 are the grouping variables.
• data is the data frame with the variable used in the formula.
• FUN is the function that will be applied to each subset.
• subset is an optional vector specifying a subset of observations to be used

举例：

aggregate(Sepal.Length ~ Species, data = ir, mean)
针对单一对象，一种分组，相当于：
aggregate(ir$Sepal.Length, by= list(ir$Species), FUN=mean)

aggregate(Sepal.Length ~ Species + indoor, data = ir, mean)
针对单一对象，多种分组，相当于：
mean_by_sp_ind = aggregate(ir$Sepal.Length, by=list(ir$Species, ir$indoor), mean)

aggregate(cbind(Sepal.Length, Sepal.Width) ~ Species, ir, mean)
针对多个对象，一种分组，相当于：
aggregate(ir[,c("Sepal.Length", "Sepal.Width")], by=list(ir$Species), mean)

针对所有对象，多个分组
aggregate(. ~ Species + indoor, data = ir, mean)

使用subset筛选满足条件的子集
aggregate(cbind(Sepal.Length, Sepal.Width) ~ Species, data = ir, subset = Petal.Width>0.6, mean)

2.3 自定义函数

• 我们可以定义一次函数，多次使用

举例：

meanX = function(vec, n){mean(head(vec[order(-vec)], n))}
aggregate(ir$Sepal.Length, by = list(ir$Species), FUN= function(x) meanX(x, 5))

• 也可以定义自己的函数

举例：

aggregate(ir$Sepal.Length, by = list(ir$Species), FUN= function(x) mean(head(x[order(-x)], 5)))

• 设置默认值

举例：

aggregate(ir$Sepal.Length, by = list(ir$Species),
FUN= function(x, n=5) mean(head(x[order(-x)], n)))

aggregate(ir$Sepal.Length, n=5, by = list(ir$Species),
FUN= function(x, n) mean(head(x[order(-x)], n)))

3. 基本的数据分析

3.2 选取样本了解数据

• View(dataset) will show the whole dataset in a new window.
• tail(dataset,x) will show the last x intances of dataset in the console.
• head(dataset,x) will show the first x instances of dataset in the console.
• names(dataset) will return the name of the attributes in the dataset.
• str(dataset) will return a summary of of the type of each attribute and the first few values.

3.3 中心性检测

1. 均值：mean()、 colMeans()、 rowMeans()
2. 中位数：median()

3.4 离散分析

1. 标准差: sd()
2. 范围: range()
3. 四分位数：IQR = Q3 − Q1
   分位数是将总体的全部数据按大小顺序排列后，处于各等分位置的变量值。如果将全部数据分成相等的两部分，它就是中位数；如果分成四等分，就是四分位数。

3.5 相关性

1. 相关性系数： cor()

4. 高级数据分析 dplyr

These are: filter(), arrange(), select(), mutate(), summarize(), sample_n(), sample_frac(), and group_by().

4.1 filter

得到满足条件的行
filter(data frame, condition).

filter(iris, Species=="setosa") # using dplyr
等效于
iris[iris$Species=="setosa",] # using basic R

4.2 arrange

arrange(data frame, attributes)
将数据帧按照属性排序

升序：
arrange(iris, Sepal.Length) #dplyr
iris[order(iris$Sepal.Length, decreasing = FALSE),] # basic R
iris[order(iris$Sepal.Length),] # basic R

降序：
arrange(iris, desc(Sepal.Length))
iris[order(iris$Sepal.Length, decreasing = TRUE),]
iris[order(-iris$Sepal.Length),]
arrange(iris, Sepal.Length,Sepal.Width)[1:5,]

升序与降序结合：
arrange(iris, Sepal.Length, desc(Sepal.Width))[1:5,]
iris[order(iris$Sepal.Length, -iris$Sepal.Width),][1:5,]

4.3 select

select(data frame, var1,. . . ,varX).
选择满足条件的列，或者去掉某几列

select(ir,Petal.Width, Species)
ir[,c("Petal.Length","Species")]

select(ir, -Species)
ir[, -c(5)]

select可以使用类似通配符的相关功能

select(ir,starts_with("Petal")) #Petal.Length and Petal.Width
select(ir, ends_with("Length")) #Sepal.Length and Petal.Length

4.4 mutate

mutate(data frame, expression) 
添加新的列到数据帧

ir = mutate(ir, DoubleSepalL = Sepal.Length*2,
PetalRatio = Petal.Length/Petal.Width)
ir$DoubleSepalL = ir$Sepal.Length*2
ir$PetalRatio = ir$Petal.Length/ir$Petal.Width

4.5 summarize、 summarise_all

summarize(data frame, function(var1,. . . ,varX))

可以调用多个函数内置函数： sd(), min(), max(), median(), sum(), cor() (correlation), n() (length of vector)_, first() (first value), last() (last value) and n_distinct() (number of distinct values in vector).

summarise(ir, avg = mean(Sepal.Length), std= sd(Sepal.Length), total=n()) 
其中n()求行数

summary(ir)

summarise_all可以处理多个对象

均值
summarise_all(ir[,1:4],mean)
四分位点
summarise_all(ir[,1:4],quantile, probs=0.75)

4.6 Sample_n

乱序、取某几行的样本

sample_n(iris,5)
iris[sample(1:nrow(iris)),][1:5,]

4.7 Sample_frac

乱序、按比例取样本

sample_frac(iris,0.01) # 取1%的样本
等效于：
iris[sample(1:nrow(iris)),][1:ceiling(nrow(iris)*0.01),]

4.8 group_by

group_by(data frame, variable)
一般与其余函数结合使用，如： summarize

summarize(group_by(ir, Species), sd(Petal.Width))
等效于：
aggregate(ir$Petal.Width, by=list(ir$Species), FUN=sd) #Base R

求相关系数
summarize(group_by(ir, Species), r=cor(Sepal.Length, Sepal.Width))

5. piping with dplyr

将一个函数的输出作为另一个函数的输入

group_by(ir, Species) %>% summarise(avg= mean(Petal.Length))
等效于：
summarise(group_by(ir,Species), r=mean(Petal.Length))

piping将多个函数可以直接结合，使用更加有效

non_virg = ir[ir$Species!="virginica", c("Petal.Length")]
sum(non_virg>3.5)
## [1] 45

#B. Using dplyr with no piping
summarise(filter(ir,Species!="virginica",Petal.Length>3.5), n())
## n()
## 1 45

#C. Using dplyr with piping
ir %>% filter(Species!="virginica", Petal.Length>3.5) %>% nrow()
## [1] 45

注意piping只能将不同的函数连接起来，不适用basing R

ir %>%
mutate(petal_w_l = Petal.Width/Petal.Length) %>%
arrange(desc(petal_w_l)) %>%
head(3) %>% select(Species, petal_w_l)

6. 练习

1. 函数、piping、which的结合使用

注意：which.max(by_spc$Sepal.Length_mean) 中的下划线‘_‘是by_spc中列的名字，并没有特殊的含义

summ = c(min = min,max = max,mean = mean,median = median, q2={function(x) quantile(x, 0.25)},q3={function(x) quantile(x, 0.75)})
by_spc=group_by(ir, Species) %>% summarise_all(summ)

by_spc
# A tibble: 3 x 25
  Species Sepal.Length_min Sepal.Width_min Petal.Length_min
  <fct>              <dbl>           <dbl>            <dbl>
1 setosa               4.3             2.3              1  
2 versic~              4.9             2                3  
3 virgin~              4.9             2.2             4.5
......

a. Which plants have a higher mean sepal length?
by_spc[which.max(by_spc$Sepal.Length_mean),]$Species

b. Which plants have the sample with the smaller petal width?
by_spc[which.min(by_spc$Petal.Width_min),]$Species

2. 获取数据帧的每一个属性的数据类型

使用sapply、class

sapply(choco, class)

3. 获取帧的mode（数据类别中的最大值）

Obtain the mode from all of the nominal attributes in the dataset.

注意 factor 使用

• 因子就是用于表示一组数据中的类别，可以记录这组数据中的类别名称及类别数目。
• 实现研究对象的分组、分类计算
  参考链接:https://www.zhihu.com/question/48472404

举例：

choco_nominal = choco[sapply(choco,{function (x) is.factor(x)})==TRUE]
sapply(choco_nominal, {function(x) names(which.max(table(x)))})

4. 多少类别

How many distinct companies have been considered? length(unique(choco$Company))

length(unique(choco$Company))

5. 将factor装换成数值类型

• 去除一些R不能理解的字符（如2,345 转换成 2345，30% 转换成 30），gsub()
• 将factor转换成number，as.numeric()

如:

Circulation2004= as.numeric(gsub(pattern = ",", replacement="", x=books$Daily.Circulation..2004))