Add check for missing index parameter to BallCover.R

[yaraskaf]

In developing some additional cover options (as discussed in #7), I noticed that level sets were being constructed multiple times per Mapper construction for the BallCover but not the FixedIntervalCover.

This seems to be because FixedIntervalCover only constructs self$level_sets when the index parameter is missing or null (i.e. when the cover is initially constructed). BallCover does not check whether index is null before constructing self$level_sets, so the computation may be performed more than once.

Addition of this check for null/missing speeds up BallCover considerably without affecting the level sets or complex that is generated.

For example, using data and filter

library(Mapper)

data("noisy_circle", package = "Mapper")
X = noisy_circle

left_pt <- X[which.min(X[, 1]),]
f_X <- matrix(apply(X, 1, function(pt) (pt - left_pt)[1]))

I ran the following code using the current version of the master branch of Mapper:

ptm <- proc.time()
base_m <- MapperRef$new(X)$
  use_filter(filter = f_X)$
  use_cover(cover="ball", epsilon=0.01)$
  use_distance_measure(measure="euclidean")$
  construct_k_skeleton(k=1L)
proc.time()-ptm

>#   user  system elapsed
     2.09    0.05    2.14

Then I ran the same code after loading the current version of this ballcover_optimization branch:

library(devtools)
load_all()

ptm <- proc.time()
opt_m <- MapperRef$new(X)$
  use_filter(filter = f_X)$
  use_cover(cover="ball", epsilon=0.01)$
  use_distance_measure(measure="euclidean")$
  construct_k_skeleton(k=1L)
proc.time()-ptm

>#   user  system elapsed
     0.23    0.00    0.19

And compared the resulting complexes and level sets:

base_levelsets <- base_m$cover$level_sets
base_complex <- base_m$simplicial_complex

optimized_levelsets <- opt_m$cover$level_sets
optimized_complex <- opt_m$simplicial_complex

expect_equal(base_complex, optimized_complex)
expect_identical(base_levelsets, optimized_levelsets)