Search very locally to find slightly better projections to polish a broader search.

search_polish(
  current,
  alpha = 0.5,
  index,
  tries,
  polish_max_tries = 30,
  cur_index = NA,
  n_sample = 100,
  polish_cooling = 1,
  ...
)

Arguments

current: the current projection basis
alpha: the angle used to search the target basis from the current basis
index: index function
tries: the counter of the outer loop of the opotimiser
polish_max_tries: maximum number of iteration before giving up
cur_index: the index value of the current basis
n_sample: number of samples to generate
polish_cooling: percentage of reduction in polish_alpha when no better basis is found
...: other arguments being passed into the search_polish()

Examples

data(t1)
best_proj <- t1[, , dim(t1)[3]]
attr(best_proj, "data") <- NULL
best_proj <- unclass(drop(best_proj))
animate_xy(
  flea[, 1:6],
  guided_tour(holes()),
    search_f = search_polish(
         polish_max_tries = 5),
  start = best_proj
)
#> Converting input data to the required matrix format.
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> Value  1.582   0.0 % better 
#> No better bases found after 25 tries.  Giving up.
#> Final projection: 
#> -0.527  0.039  
#> -0.040  0.391  
#> -0.037  0.456  
#> 0.059  0.524  
#> -0.846  -0.007  
#> 0.027  0.602  
#> Using half_range 66