AFLW calc_scags_wide() computation time and errors due to interp package

[harriet-mason]

The AFLW example from the UseR! talk and working paper no longer runs. The code for the example is below.

library(fitzRoy)
library(dplyr)
library(interp)
library(cassowaryr)

# Get data and find scatter plot that is instant and long scatter plot
aflw <- fetch_player_stats(2020, comp = "AFLW")

aflw_num <- aflw %>%
  select_if(is.numeric)

aflw_num <- aggregate(aflw_num[,5:37],
                      list(aflw$player.player.player.surname),
                      mean)

#This calculation used in the paper and package will take an estimated 4h and return an error after 30mins
AFLW_scags <- calc_scags_wide(aflw_num[,c(2:34)]) 

There are two issues with this calculation.

1. The computation time has been increased from approximately 12 minutes to 4 hours. This is due to a handful of plots that were previously instantaneous when calculated under tripack now taking over a minute each. There isn't a clear visual feature that is causing this issue and the only way to find out a plot has a large number of edges is to run the Delaney triangulation. Tracing back the source of the sudden increase of time we have:
   cassowary::scree > alphahull::delvor > interp::tri.mesh > interp:::shull.deltri
   The function interp::tri.mesh is a replacement for tripack::trimesh (this is where the packages differ) BUT inerp::shell.dultri isn’t written with R code, it is a wrapper for a function in C++. Therefore, even though the shull library is open source, it being written in C++ makes these errors hard to fix. An example of a scatter plot that was previously instantaneous and now takes over a minute is below.

# three vectors (one is problems)
a <- aflw_num$goals
b <- aflw_num$behinds
c <- aflw_num$handballs
# a,b is fast, a,c is slow (1.5mins)

# get ac to shull input
#inside cassowaryr::scree()
ac <- cbind(cassowaryr:::unitize(a), cassowaryr:::unitize(c))
dupes <- paste(a, c, sep =",")
ac <- ac[!duplicated(dupes),]

#inside alphahull::delvor()
y=NULL
AC <- xy.coords(ac, y)
tri.obj1 <- interp::tri.mesh(AC) #1min 20seconds
tri.obj2 <- tripack::tri.mesh(AC) #instantaneous

2. There is an "error counting arcs" that I cannot preemptively check for that was not returned with tripack.

# Scatter plot that throws an error
d <- aflw_num$totalPossessions
e <- aflw_num$contestedPossessions
scree(d,e) # returns an error from the shull.deltri function

Since the error comes from the C++ code, I can't debug it.