[R-pkg-devel] How to get arbitrary precise inputs from R for an Rcpp package?

[Duncan Murdoch]

Just for your info:

 > exp(mpfr("0.1", 152))
1 'mpfr' number of precision  152   bits
[1] 1.1051709180756476248117078264902466682245471948

So it looks like your approach of parsing the numbers yourself makes sense.

You can find the R parser here: 
https://github.com/wch/r-source/blob/b64422334a8269535718efd9a1f969c94b103056/src/main/gram.y#L2577-L2705

I don't know how much of that you want to replicate, but I suppose 
handling the weird cases (e.g. 0x1p1) the way R does will make it easier 
for your users.

Duncan Murdoch

On 2024-07-18 4:29 p.m., Khue Tran wrote:
> Hi,
> 
> I am trying to create an Rcpp package that involves arbitrary precise
> calculations. The function to calculate e^x below with 100 digits precision
> works well with integers, but for decimals, since the input is a double,
> the result differs a lot from the arbitrary precise result I got on
> Wolfram.
> 
> I understand the results are different since 0.1 cannot be represented
> precisely in binary with limited bits. It is possible to enter 1 then 10
> and get the multiprecision division of these two integers to attain a more
> precise 0.1 in C++, but this method won't work on a large scale. Thus, I am
> looking for a general solution to get more precise inputs?
> 
> The dummy example is as follows:
> 
> library(Rcpp)
> 
> sourceCpp(code = '
> #include <boost/multiprecision/cpp_dec_float.hpp>
> #include <boost/math/special_functions/expm1.hpp>
> 
> // [[Rcpp::depends(BH)]]
> // [[Rcpp::export]]
> std::string calculateExp100(double x) {
>    boost::multiprecision::cpp_dec_float_100 bx = x;
>    boost::multiprecision::cpp_dec_float_100 expx = exp(bx);
>    return expx.str(50);
> }
> ')
> 
> [1] represents the R output and [W] for Wolfram results
> 
> calculateExp100(1) # Agrees with Wolfram answer all the way to the last
> digit
> [1] "2.7182818284590452353602874713526624977572470937"
> [W] 2.7182818284590452353602874713526624977572470936999595749669676277...
> 
> calculateExp100(0.1) # Differs pretty significantly from Wolfram's answer
> [1] "1.1051709180756476309466388234587796577416634163742"
> [W] 1.1051709180756476248117078264902466682245471947375187187928632894...
> 
> I am currently trying to get precise inputs by taking strings instead of
> numbers then writing a function to decompose the string into a rational
> with the denominator in the form of 10^(-n) where n is the number of
> decimal places. I am not sure if this is the only way or if there is a
> better method out there that I do not know of, so if you can think of a
> general way to get precise inputs from users, it will be greatly
> appreciated!
> 
> Thank you!
> Best,
> Khue Tran
>