We modified the function dist_make() to pass along additional arguments to
the distance function. This helps with distance functions that take extra
parameters. We removed the method keyword argument from dist_make() because
we found it not to be useful in practice.
We added a new function, dist_multi_centroid(), to produce a distance matrix
between multiple centroid positions.
We fixed an error in dist_groups() that will pop up in future versions of R.
In R 4.0.0, the data.frame() function will have a new default value,
stringsAsFactors = FALSE. This update caused one column in the output
from dist_groups() to change from a factor to a character vector, which broke
one of our tests. We updated the code to deliberately make this column a
factor. The function's behavior will be preserved when R 4.0.0 is released.
We've made two major updates for this release.
The centroid functions have gained a keyword argument, squared. The final
step in computing distance to group centroids involves taking a square root.
Sometimes, we end up with a negative number inside the square root. Normally,
this produces NaN as a result. However, if squared is set to TRUE, we
don't take the square root and the result is always a real number. The default
setting is squared = FALSE, which gives the distance as you'd expect. Thanks
to Sam Ross for helpful advice on this topic.
We added a new function, pivot_to_numeric_matrix(). This function takes a
data frame in long format and converts to a matrix suitable for distance
calculations. Long-format data frames are commonly used with functions in the
tidyverse, and proper conversion to a matrix requires a few non-obvious
steps. The packages dplyr, tidyr, and tibble are needed to run the
function, and have been added as suggested packages for usedist.
During development, we had implemented an additional function to create a
distance matrix directly from a data frame in long format. However, we found
that it was nearly as convenient to use pivot_to_numeric_matrix() and
dist_make() together to achieve the same result. We added an example to the
README file to illustrate this.
Initial release!
For the first release, we've moved over all the distance matrix-related
functions from the qiimer package. Utility functions dist_setNames and
dist_make were added to the existing functions dist_subset, dist_get, and
dist_groups.
The other area of focus for our initial release was a set of distance-to-centroid functions. These functions operate on a set of items that are placed into several groups, and they apply algebraic formulas to find the group centroid distances based on the distance matrix. If you had a set of points in normal Euclidean space, you could easily find the centroid positions and compute the distances from there. The centroid functions included in the package are more useful for non-Euclidean distances, where it is less straightforward to locate the centroids in space. The centroid functions were inspired by work in microbial ecology, where it is typical to work with non-Euclidean distances between samples.