Vectorization

[rstub]

Currently only day_of_week is vectorized. Methods like swe_calc or swe_fixstar take two arguments that could be used for vectorization: the date and the object. From my point of view one cannot rule out either of these arguments for vectorization. So for a simple interface that stays close to the C API I can imagine two extensions. In both cases the returned coordinates will form a NumericMatrix instead of a NumericVector and both date and object are of appropriate vector type:

1. Only one of the arguments date and object is allowed to have length > 1. So one can either compute many objects for one date or one object for many dates.

2. Both arguments date and object must have the same length. The n. position corresponds to the n. date and n. object.

@vreijs Which form do you think is most useful?

[vreijs]

In ARCHOAECOSMO I use the 2) bullet but it also does 1. It uses the general R rule that if a smaller vector can be multiple times within the large vector the smaller is repeated (only when length(vector1)/length(vector) is an integer). This is automatically done by using data_frame(), for instance: functionvector <- data.frame( JDNDaysUT, Lat, Longitude, HeightEye, TempE, PresE, ObjectName, EclipseType, AppAlt, EventPhase, stringsAsFactors = FALSE ) If all input vectors have only one member, will the results be the same as we get now? That is what my method also does. Would be nice if that would also be the case with swe functions. In that case the results are still the same as swe.

[rstub]

See the branch https://github.com/rstub/swephR/tree/feature/vectorization for a sample implementation. The present interface stays the same, but it is also possible to use vectors of bodies or dates together with swe_calc.

If this works for you, I can do the same with other methods.

[vreijs]

Hello Ralk,

My method is like this (I use this method for vectorising all my ARCHAEOCOSMO functions):

###################################################################
swe_calc <- function (Date,
                          ipl,
                          flag) {
  functionvector <-
    data.frame(Datemipl,flag, stringsAsFactors = FALSE)
  
  #    print(functionvector)
  ResultVector <- c(0)
  for (i in 1:nrow(functionvector))
  {
    ResultVector[i] = S_calc(  #S meaning single value function
      functionvector$Date[i],
      functionvector$ipl[i],
      functionvector$flag[i]
    )
  }
  return(ResultVector)
}

[vreijs]

How do I download this branch... So I went to your link and then I pressed 'Clone or Donwload' button in GitHub, this gives me the link: https://github.com/rstub/swephR.git But is that the correct branch? So how do I refer this this branch in RStudio?

…

On 1 September 2018 at 10:40, Ralf Stubner ***@***.***> wrote: See the branch https://github.com/rstub/swephR/tree/feature/vectorization for a sample implementation.

[rstub]

Two possibilities:

• In RStudio go to the 'Git' tab: In the upper right corner you can select the branch: use 'feature/vectorization' and press 'build & install'

• Use devtools::install_github("rstub/swephR@feature/vectorization") to install directly from the GH branch

[vreijs]

On 3 September 2018 at 21:56, Ralf Stubner ***@***.***> wrote: - In RStudio go to the 'Git' tab: In the upper right corner you can select the branch: use 'feature/vectorization' and press 'build & install'

Thanks that worked. Looks good. Lets go with that for all functions.

[vreijs]

We sill need to pick this vectorisation up. I will try to make an evaluation of our two methods.