使用精确匹配和模糊匹配连接R中的两个大型数据集
我正在尝试内部连接两个数据集:使用精确匹配和模糊匹配连接R中的两个大型数据集,r,join,inner-join,fuzzyjoin,R,Join,Inner Join,Fuzzyjoin,我正在尝试内部连接两个数据集:df150000个obs,如下所示: Name | Line.1 | Line.2 | Town | County | Postcode -------------------|------------------|------------|------------|--------------|---------- ACME Inc | 63 Long S
df1
50000个obs,如下所示:
Name | Line.1 | Line.2 | Town | County | Postcode
-------------------|------------------|------------|------------|--------------|----------
ACME Inc | 63 Long Street | | Fakeington | Lincolnshire | PA4 8QU
BETA LTD | 91a | Main Drove | Cloud City | Something | BN1 6LD
The Giga | 344 Lorem Street | | Ipsom | Dolor | G2 8LY
Name | AddressLine1 | AddressLine2 | AddressLine3 | AddressLine4 | Postcode | RatingValue
-------------------|----------------|------------------|--------------|--------------|----------|-------------
ACME | | 63 Long Street | Fakeington | Lincolnshire | PA4 8QU | 1
Random Company | | Rose Ave | Fakeington | | AB2 51GL | 5
BETA Limited | Business House | 91a Main Drove | Something | | BN1 6LD | 3
Giga Incorporated | | 344 Lorem Street | Ipsum | Dolor | G2 8LY | 5
df2
500000 obs的外观如下所示:
Name | Line.1 | Line.2 | Town | County | Postcode
-------------------|------------------|------------|------------|--------------|----------
ACME Inc | 63 Long Street | | Fakeington | Lincolnshire | PA4 8QU
BETA LTD | 91a | Main Drove | Cloud City | Something | BN1 6LD
The Giga | 344 Lorem Street | | Ipsom | Dolor | G2 8LY
Name | AddressLine1 | AddressLine2 | AddressLine3 | AddressLine4 | Postcode | RatingValue
-------------------|----------------|------------------|--------------|--------------|----------|-------------
ACME | | 63 Long Street | Fakeington | Lincolnshire | PA4 8QU | 1
Random Company | | Rose Ave | Fakeington | | AB2 51GL | 5
BETA Limited | Business House | 91a Main Drove | Something | | BN1 6LD | 3
Giga Incorporated | | 344 Lorem Street | Ipsum | Dolor | G2 8LY | 5
我想谈谈类似于df_final
Name | Postcode | RatingValue
-------------------|----------|-------------
ACME Inc | PA4 8QU | 1
BETA LTD | BN1 6LD | 3
Giga Incorporated | G2 8LY | 5
这些是一对一的匹配,df1
中的所有值都应该存在于df2
中Postcode
是一个精确匹配,而地址被分割成多行,没有规则的模式,因此我认为我最好的选择是通过Name
匹配
我尝试了fuzzyjoin
软件包,但是我得到了一个错误:无法分配大小为120.6GB的向量,因此我想我必须使用另一种方法来处理更大的数据集
你有没有什么好办法
df1 <- data.frame(
stringsAsFactors = FALSE,
Name = c("ACME Inc", "BETA LTD", "Giga Incorporated"),
Line.1 = c("63 Long Street", "91a", "344 Lorem Street"),
Line.2 = c(NA, "Main Drove", NA),
Town = c("Fakeington", "Cloud City", "Ipsom"),
County = c("Lincolnshire", "Something", "Dolor"),
Postcode = c("PA4 8QU", "BN1 6LD", "G2 8LY")
)
df2 <- data.frame(
stringsAsFactors = FALSE,
Name = c("ACME", "Random Company","BETA Limited","Giga Incorporated"),
AddressLine1 = c(NA, NA, "Business House", NA),
AddressLine2 = c("63 Long Street", "Rose Ave","91a Main Drove","344 Lorem Street"),
AddressLine3 = c("Fakeington", "Fakeington", "Something", "Ipsum"),
AddressLine4 = c("Lincolnshire", NA, NA, "Dolor"),
Postcode = c("PA4 8QU", "AB2 51GL", "BN1 6LD", "G2 8LY"),
RatingValue = c(1L, 5L, 3L, 5L)
)
df1也许像下面这样的东西可以满足问题的要求。它使用包stringdist
,而不是fuzzyjoin
首先,merge
byPostcode
,因为匹配是精确的。然后获取名称
之间的相似性。如果它们高于预定阈值,则保留这些行
thresh <- 0.75
df_final <- merge(df2[c(1, 6:7)], df1[c(1, 6)], by = "Postcode", suffixes = c("",".y"))
i <- apply(df_final[c(2, 4)], 1, function(x) {stringdist::stringsim(x[1], x[2], method = 'jw')}) >= thresh
df_final <- df_final[i, c(2, 1, 3)]
df_final
# Name Postcode RatingValue
#1 BETA Limited BN1 6LD 3
#2 Giga Incorporated G2 8LY 5
#3 ACME PA4 8QU 1
thresh我首先将其分解为一个较小的问题。从两个现有列中创建一个数据帧:df1$Name和df2$Name,其中的行表示满足某个匹配阈值(soundex、字符串距离,任何您想要的)的“匹配”公司名称的每个组合。对df1和df2进行笛卡尔连接,然后进行过滤,以便只保留有效的组合。@BillO'Brien笛卡尔连接df1
是50K行,df2
500K,50000*500000
是2.5e+10
。我被纠正了。这似乎已经完成了。我会做一些清理来移除stopwords,调整阈值并测试不同的方法,但到目前为止还不错!