突破R内存限制的企业级大数据挖掘利器：Microsoft R Server 快速上手

R语言是一款非常优秀的数据挖掘工具，拥有顶尖的数据处理、数据挖掘课数据可视化。是数据从业者必备的一把利器。但是其基于内存的诟病也一直被人所嫌弃，虽然这几年很多优秀的扩展包极大提升了R语言的性能，但是在面对企业级大数据挖掘面前，也会显得力不从心。现在我们也不用担心R语言这个问题了，自从微软收购了商业版R以后，就进行了很多的整合和优化，之前只面向高校学生免费试用，现在，我们企业界的数据从业者也可以免费下载Microsoft R Server ，利用MRS处理大数据，MRS对开源R100%兼容，能充分利用CRAN 现有的10000+扩展包，实现不同数据挖掘需求。关于Microsoft R Server的安装，陈堰平老师已经写了一篇非常详细的文档，感兴趣者可以点击以下链接，查看文档进行安装。注意：请确保最后你都进入页面https://my.visualstudio.com，选择最新的Microsoft R Server 9.10版本下载。贴心的微软为我们提供了不同系统的安装版本，如下图所示：


我们按照堰平老师的安装步骤安装成功后，会在你的计算机出现：


说明MRS已经在你计算机安装成功。我们点击RGui，可以出现类似R相似界面：


我们也可以利用RStudio来调用MRS。操作如下：


选择MRS即可。好了，既然我们安装好了MRS，那么接下来就用一个简单的案例来帮助大家快速上手。如果数据集不大，可以直接导入到R或MRS中再进行处理（MRS速度优于R）。我们导入ccFraud.csv数据集，有一千万条记录。对比MRS和R导入数据的时间如下：



可见，利用MRS导数据的时间花了31秒，利用R导数据的时间花了1分钟，速度差不多提高了一倍。如果是大数据集，MRS也提供了将数据集先保存为.xdf格式（在硬盘中），该数据对象可供大多数RevoScaleR包中的函数使用（数据处理、数据转换、数据建模等）。我们可以利用rxImport函数实现，将其outFile参数设置为你要保存的文件名即可。

> # 导入csv数据集
> readpath <- "D:/MRS/Data"
> infile <- file.path(readpath,"ccFraud.csv")
> ccFraud_xdf <- rxImport(inData = infile,
+                         outFile = "ccFraud.xdf",
+                         overwrite = TRUE)
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 1.393 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 1.440 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 1.462 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 1.462 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 1.475 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 1.408 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 1.454 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 1.381 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 1.417 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 1.429 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 1.440 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 1.425 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 1.452 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 1.456 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 1.406 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 1.379 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 1.434 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 1.409 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 1.422 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 1.384 seconds
>

结束后，我们利用rxGetInfo函数查看.xdf文件的数据结构（将参数getVarInfo设置为TRUE），并查看数据的前十行（numRows = 10）。

> rxGetInfo("ccFraud.xdf",getVarInfo = TRUE,numRows = 10)
File name: C:\Program Files\Microsoft\R Server\R_SERVER\library\RevoScaleR\rxLibs\x64\ccFraud.xdf
Number of observations: 1e+07
Number of variables: 9
Number of blocks: 20
Compression type: zlib
Variable information:
Var 1: custID, Type: integer, Low/High: (1, 1e+07)
Var 2: gender, Type: integer, Low/High: (1, 2)
Var 3: state, Type: integer, Low/High: (1, 51)
Var 4: cardholder, Type: integer, Low/High: (1, 2)
Var 5: balance, Type: integer, Low/High: (0, 41485)
Var 6: numTrans, Type: integer, Low/High: (0, 100)
Var 7: numIntlTrans, Type: integer, Low/High: (0, 60)
Var 8: creditLine, Type: integer, Low/High: (1, 75)
Var 9: fraudRisk, Type: integer, Low/High: (0, 1)
Data (10 rows starting with row 1):
custID gender state cardholder balance numTrans numIntlTrans creditLine fraudRisk
1       1      1    35          1    3000        4           14          2         0
2       2      2     2          1       0        9            0         18         0
3       3      2     2          1       0       27            9         16         0
4       4      1    15          1       0       12            0          5         0
5       5      1    46          1       0       11           16          7         0
6       6      2    44          2    5546       21            0         13         0
7       7      1     3          1    2000       41            0          1         0
8       8      1    10          1    6016       20            3          6         0
9       9      2    32          1    2428        4           10         22         0
10     10      1    23          1       0       18           56          5         0
>

我们也可以在保存.xdf文件时，利用stringsAsFactors，colClasses，和colInfo等参数改变变量的数据类型。比如我们利用colInfo将变量gender从数值型变为因子型，且因子水平为“F”、“M”，利用colClasses将变量fraudRisk从数值型变成因子型。

> # 改变变量的数据存储类型
> ccFraud_xdf <- rxImport(inData = infile,
+                         outFile = "ccFraud.xdf",
+                         colClasses = c(fraudRisk = "factor"),
+                         colInfo = list("gender" = list(type = "factor",
+                                                        levels = c("1","2"),
+                                                        newLevels = c("F","M"))),
+                         overwrite = TRUE)
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 1.871 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 1.832 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 1.825 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 1.808 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 2.018 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 2.061 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 2.158 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 1.917 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 1.852 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 1.795 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 1.829 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 1.793 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 1.849 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 1.806 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 1.773 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 1.813 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 1.812 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 1.850 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 1.824 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 1.828 seconds
>
> # 查看ccFraud_xdf的数据结构
> rxGetInfo(ccFraud_xdf,getVarInfo = TRUE,numRows = 5)
File name: C:\Program Files\Microsoft\R Server\R_SERVER\library\RevoScaleR\rxLibs\x64\ccFraud.xdf
Number of observations: 1e+07
Number of variables: 9
Number of blocks: 20
Compression type: zlib
Variable information:
Var 1: custID, Type: integer, Low/High: (1, 1e+07)
Var 2: gender
2 factor levels: F M
Var 3: state, Type: integer, Low/High: (1, 51)
Var 4: cardholder, Type: integer, Low/High: (1, 2)
Var 5: balance, Type: integer, Low/High: (0, 41485)
Var 6: numTrans, Type: integer, Low/High: (0, 100)
Var 7: numIntlTrans, Type: integer, Low/High: (0, 60)
Var 8: creditLine, Type: integer, Low/High: (1, 75)
Var 9: fraudRisk
2 factor levels: 0 1
Data (5 rows starting with row 1):
custID gender state cardholder balance numTrans numIntlTrans creditLine fraudRisk
1      1      F    35          1    3000        4           14          2         0
2      2      M     2          1       0        9            0         18         0
3      3      M     2          1       0       27            9         16         0
4      4      F    15          1       0       12            0          5         0
5      5      F    46          1       0       11           16          7         0
>

从数据结构可知，变量的类型已经发生改变，且gender的因子水平从1、2变成F、M。我们也可以对.xdf文件进行描述性统计分析，通过rxSummary函数实现。

> # 利用rxSummary函数对数据进行描述性统计分析
> rxSummary(~.,ccFraud_xdf) # 对全部变量进行统计
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.069 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.072 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.078 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.079 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.081 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.081 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.077 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.085 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.078 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.082 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.084 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.082 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.079 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.086 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.078 seconds
Computation time: 1.661 seconds.
Call:
rxSummary(formula = ~., data = ccFraud_xdf)

Summary Statistics Results for: ~.
Data: ccFraud_xdf (RxXdfData Data Source)
File name: ccFraud.xdf
Number of valid observations: 1e+07

Name         Mean         StdDev       Min Max      ValidObs MissingObs
custID       5.000001e+06 2.886751e+06 1   10000000 1e+07    0
state        2.466127e+01 1.497012e+01 1         51 1e+07    0
cardholder   1.030004e+00 1.705991e-01 1          2 1e+07    0
balance      4.109920e+03 3.996847e+03 0      41485 1e+07    0
numTrans     2.893519e+01 2.655378e+01 0        100 1e+07    0
numIntlTrans 4.047190e+00 8.602970e+00 0         60 1e+07    0
creditLine   9.134469e+00 9.641974e+00 1         75 1e+07    0

Category Counts for gender
Number of categories: 2
Number of valid observations: 1e+07
Number of missing observations: 0

gender Counts
F      6178231
M      3821769

Category Counts for fraudRisk
Number of categories: 2
Number of valid observations: 1e+07
Number of missing observations: 0

fraudRisk Counts
0         9403986
1          596014
>

跟普通summary函数相似，对数值型变量返回平均值、标准差、最小值、最大值、样本个数和缺失值个数，对因子型变量则返回频数。除了这些简单的处理外，ScaleR也包含了丰富的数据处理和算法，具体如下所示：


最后，让我们利用rxLogit函数构建Logistic回归模型（R中的glm函数也适用），并利用summary函数查看模型信息。

> # logitic回归模型
> ccFraudglm <- rxLogit(fraudRisk ~ gender + cardholder + balance + numTrans
+                       + numIntlTrans + creditLine,data = ccFraud_xdf)
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.081 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.080 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.071 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.071 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.068 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.075 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.070 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.075 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.069 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.073 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.072 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.067 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.077 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.074 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.073 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.070 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.077 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.069 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.071 seconds

Starting values (iteration 1) time: 1.558 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.061 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.182 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.184 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.197 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.200 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.185 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.186 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.196 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.192 seconds

Iteration 2 time: 3.866 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.070 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.189 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.205 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.194 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.194 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.194 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.187 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.184 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.201 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.208 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.216 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.191 seconds

Iteration 3 time: 3.950 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.068 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.197 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.187 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.201 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.211 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.189 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.198 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.181 seconds

Iteration 4 time: 3.914 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.061 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.197 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.201 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.189 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.204 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.196 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.202 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.194 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.185 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.207 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.186 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.202 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.209 seconds

Iteration 5 time: 3.963 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.066 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.198 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.187 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.200 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.189 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.197 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.187 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.197 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.187 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.189 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.208 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.204 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.184 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.186 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.197 seconds

Iteration 6 time: 3.903 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.061 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.198 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.198 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.204 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.196 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.201 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.197 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.181 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.193 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.198 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.194 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.192 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.201 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.198 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.201 seconds

Iteration 7 time: 3.956 secs.
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.060 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.189 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.199 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.215 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.231 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.220 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.202 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.213 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.211 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.191 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.188 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.209 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.190 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.195 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.202 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.222 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.224 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.213 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.223 seconds

Iteration 8 time: 4.194 secs.

Elapsed computation time: 29.316 secs.
> # 查看模型结果
> summary(ccFraudglm)
Call:
rxLogit(formula = fraudRisk ~ gender + cardholder + balance +
numTrans + numIntlTrans + creditLine, data = ccFraud_xdf)

Logistic Regression Results for: fraudRisk ~ gender + cardholder + balance +
numTrans + numIntlTrans + creditLine
Data: ccFraud_xdf (RxXdfData Data Source)
File name: ccFraud.xdf
Dependent variable(s): fraudRisk
Total independent variables: 8 (Including number dropped: 1)
Number of valid observations: 1e+07
Number of missing observations: 0
-2*LogLikelihood: 2149329.7462 (Residual deviance on 9999993 degrees of freedom)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept)  -8.888e+00  1.292e-02 -687.88 2.22e-16 ***
gender=F     -6.010e-01  3.716e-03 -161.73 2.22e-16 ***
gender=M        Dropped    Dropped Dropped  Dropped
cardholder    4.703e-01  9.749e-03   48.24 2.22e-16 ***
balance       3.755e-04  4.558e-07  823.71 2.22e-16 ***
numTrans      4.659e-02  6.526e-05  713.86 2.22e-16 ***
numIntlTrans  2.967e-02  1.757e-04  168.83 2.22e-16 ***
creditLine    9.297e-02  1.389e-04  669.13 2.22e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Condition number of final variance-covariance matrix: 5.5717
Number of iterations: 8
>

模型建好后，可以利用rxPredict函数可以对数据进行预测。

> # 利用logit模型进行预测
> rxPredict(ccFraudglm,data = ccFraud_xdf,
+           outData = ccFraud_xdf, overwrite=TRUE)
Rows Read: 500000, Total Rows Processed: 500000, Total Chunk Time: 0.116 seconds
Rows Read: 500000, Total Rows Processed: 1000000, Total Chunk Time: 0.271 seconds
Rows Read: 500000, Total Rows Processed: 1500000, Total Chunk Time: 0.284 seconds
Rows Read: 500000, Total Rows Processed: 2000000, Total Chunk Time: 0.268 seconds
Rows Read: 500000, Total Rows Processed: 2500000, Total Chunk Time: 0.270 seconds
Rows Read: 500000, Total Rows Processed: 3000000, Total Chunk Time: 0.268 seconds
Rows Read: 500000, Total Rows Processed: 3500000, Total Chunk Time: 0.297 seconds
Rows Read: 500000, Total Rows Processed: 4000000, Total Chunk Time: 0.269 seconds
Rows Read: 500000, Total Rows Processed: 4500000, Total Chunk Time: 0.271 seconds
Rows Read: 500000, Total Rows Processed: 5000000, Total Chunk Time: 0.273 seconds
Rows Read: 500000, Total Rows Processed: 5500000, Total Chunk Time: 0.302 seconds
Rows Read: 500000, Total Rows Processed: 6000000, Total Chunk Time: 0.287 seconds
Rows Read: 500000, Total Rows Processed: 6500000, Total Chunk Time: 0.270 seconds
Rows Read: 500000, Total Rows Processed: 7000000, Total Chunk Time: 0.279 seconds
Rows Read: 500000, Total Rows Processed: 7500000, Total Chunk Time: 0.291 seconds
Rows Read: 500000, Total Rows Processed: 8000000, Total Chunk Time: 0.271 seconds
Rows Read: 500000, Total Rows Processed: 8500000, Total Chunk Time: 0.284 seconds
Rows Read: 500000, Total Rows Processed: 9000000, Total Chunk Time: 0.305 seconds
Rows Read: 500000, Total Rows Processed: 9500000, Total Chunk Time: 0.280 seconds
Rows Read: 500000, Total Rows Processed: 10000000, Total Chunk Time: 0.278 seconds
>
> # 查看预测结果的前十条记录
> rxGetInfo(ccFraud_xdf,numRows = 10)
File name: C:\Program Files\Microsoft\R Server\R_SERVER\library\RevoScaleR\rxLibs\x64\ccFraud.xdf
Number of observations: 1e+07
Number of variables: 10
Number of blocks: 20
Compression type: zlib
Data (10 rows starting with row 1):
custID gender state cardholder balance numTrans numIntlTrans creditLine fraudRisk fraudRisk_Pred
1       1      F    35          1    3000        4           14          2         0   0.0008208775
2       2      M     2          1       0        9            0         18         0   0.0017884961
3       3      M     2          1       0       27            9         16         0   0.0044740019
4       4      F    15          1       0       12            0          5         0   0.0003372351
5       5      F    46          1       0       11           16          7         0   0.0006229633
6       6      M    44          2    5546       21            0         13         0   0.0246553959
7       7      F     3          1    2000       41            0          1         0   0.0018993247
8       8      F    10          1    6016       20            3          6         0   0.0055909255
9       9      M    32          1    2428        4           10         22         0   0.0068448406
10     10      F    23          1       0       18           56          5         0   0.0023438135

在数据的最后一列增加预测结果。好了，今晚就先分享到这里，目的是让大家了解Microsoft R Server的一些基本用法。如果大家面临企业大数据难以分析建模的困境，可以下载安装MRS来尝试解决你们现实的业务问题。除了以上简单函数意外，微软也专门开发了一个MicrosoftML包，其提供了新的机器学习功能，具有更高的速度，性能和可扩展性，特别是处理大量的文本数据或高维分类数据。


感兴趣的读者可以去查看相关帮助文档。