不同的data.table r结果有或没有引用/如何使用uniqueN计算唯一值

Question

I'm trying to select all claims (ICN) with the value of 7015061009422. I use the following code:

  dt[ICN==7015061009422]

And I get back

> dt[ICN==7015061009422]
    UniversalID           ICN
 1:           2 7015061009422
 2:           3 7015061009417
 3:           2 7015061009411
 4:           2 7015061009428
 5:           2 7015061009437
 6:           4 7015061009417
 7:           5 7015061009411
 8:           6 7015061009417
 9:           6 7015061009422
10:           7 7015061009422

I finally figured out that if I put quotations around the value, what I want to happen, which is to select all the lines that actually have the ICN number equal to 7015061009422, that is what I get:

    > dt[ICN=="7015061009422"]
   UniversalID           ICN
1:           2 7015061009422
2:           6 7015061009422
3:           7 7015061009422

Why is it that using quotations around my data values makes such a big difference?

Sample dataset:

    ICN<-c(7015061009422,7015061009417,7015061009411,7015061009428,7015061009437,7,7015061009417,7015061009411,7015061009417,7015061009422,7015061009422,1)
  UniversalID<-c(2,3,2,2,2,1,4,5,6,6,7,8)
  dt<-cbind(UniversalID,ICN)
  dt<-as.data.table(dt)
  dt[ICN==7015061009422]
  dt[ICN=="7015061009422"]

I'm hoping that this can also help me figure out why my unique count isn't working:

> > dt[,uniqueN(ICN)] 
[1] 3

Very clearly, there are more than three different ICN values, so why this happening?

Answer 1

When you compare very large numbers using == , R does not check for exact equality. Rather, it checks whether the numbers are "close" to each other. If so, it returns TRUE. R does the same thing with floating point numbers.

When you put quotations around your number, you are implicitly casting it into a string. Thus, calling == will cause R to do a string comparison, and you get the answer you want.

Instead of a == b , use abs(ab) < 0.001 , or some other arbitrary small number.

EDIT: Are you doing any actual arithmetic with these numbers? If not, you should probably just convert them to factors.