This is a novel method for finding more interesting projections for the
guided tour. It works by first taking a small step in n random
directions, and then picking the direction that looks most promising
(based on the height of the index function), which is effectively a gradient search.
Then it performs a linear search along the geodesic in that direction,
traveling up to half way around the sphere.
Usage
search_geodesic(
current,
alpha = 1,
index,
tries,
max.tries = 5,
...,
n = 5,
delta = 0.01,
cur_index = NA
)Arguments
- current
starting projection
- alpha
maximum distance to travel (currently ignored)
- index
interestingness index function
- tries
the counter of the outer loop of the opotimiser
- max.tries
maximum number of failed attempts before giving up
- ...
other arguments being passed into the
search_geodesic()- n
number of random steps to take to find best direction
- delta
step size for evaluation of best direction
- cur_index
index value for starting projection, set NA if it needs to be calculated
Details
You should not to have call this function directly, but should supply it
to the guided_tour as a search strategy.
Examples
animate_xy(flea[, 1:6], guided_tour(holes(), search_f = search_geodesic))
#> Converting input data to the required matrix format.
#> Target: 0.629, 68.5% better
#> Using half_range 4.4
