mean.integer64 method could return numeric

[jangorecki]

Hello,
Base R mean on integers will always return numeric. Would you consider aligning this behavior in bit64?

[truecluster]

Jan,
integer64() are deliberately different from integer and from double:

`

require(bit64)

consider a value that double cannot count anymore

x <- lim.integer64()[2]

this mean is exact

mean(c(x, x-2))
integer64
[1] 9223372036854775806

well, as exact as it can be

mean(c(x, x-1))
integer64
[1] 9223372036854775806

similar with quantiles

quantile(x-(0:100))
integer64
0% 25% 50% 75% 100%
9223372036854775707 9223372036854775732 9223372036854775757 9223372036854775782 9223372036854775807

note that these are meaningful distinguishable quantiles

unique(quantile(x-(0:100)))
integer64
[1] 9223372036854775707 9223372036854775732 9223372036854775757 9223372036854775782 9223372036854775807

if you want numeric results you always can

mean(as.double(c(x,x-2)))
[1] 9.223372e+18
Warnmeldung:
In as.double.integer64(c(x, x - 2)) :
integer precision lost while converting to double

or

as.double(mean(c(x,x-2)))
[1] 9.223372e+18
Warnmeldung:
In as.double.integer64(mean(c(x, x - 2))) :
integer precision lost while converting to double

but note that you cannot undo the precision loss

as.integer64(mean(as.double(c(x,x-2))))
integer64
[1]
Warnmeldung:
In as.double.integer64(c(x, x - 2)) :
integer precision lost while converting to double

actually the implementation returns NAa above ~ 2^52

similar with quantiles

quantile(as.double(x-(0:100)))
0% 25% 50% 75% 100%
9.223372e+18 9.223372e+18 9.223372e+18 9.223372e+18 9.223372e+18
Warnmeldung:
In as.double.integer64(x - (0:100)) :
integer precision lost while converting to double
as.double(quantile(x-(0:100)))
[1] 9.223372e+18 9.223372e+18 9.223372e+18 9.223372e+18 9.223372e+18
Warnmeldung:
In as.double.integer64(quantile(x - (0:100))) :
integer precision lost while converting to double

note that using double all the quantiles are the same (not meaningful)

unique(quantile(as.double(x-(0:100))))
[1] 9.223372e+18
Warnmeldung:
In as.double.integer64(x - (0:100)) :
integer precision lost while converting to double
`

[jangorecki]

Thanks for prompt reply. I now realized that

mean(as.integer64(c(3,4)))

is not returning 3.5 but a whole number. Therefore there is no reason to return numeric type when it always returns whole numbers.