Hyperoverlap can be used to detect and visualise overlap in n-dimensional space.
To explore the functions in hyperoverlap, we’ll use the iris dataset. This dataset contains 150 observations of three species of iris (“setosa”, “versicolor” and “virginica”). These data are four-dimensional (Sepal.Length, Sepal.Width, Petal.Length, Petal.Width) and are documented in ?iris. We’ll set up five test datasets to explore the different functions: 1. test1 two entities (setosa, virginica); three dimensions (Sepal.Length, Sepal.Width, Petal.Length) 1. test2 two entities (versicolor, virginica); three dimensions (as above) 1. test3 two entities (setosa, virginica); four dimensions 1. test4 two entities (versicolor, virginica); four dimensions 1. test5 all entities, all dimensions
test1 <- iris[which(iris$Species!="versicolor"),c(1:3,5)]
test2 <- iris[which(iris$Species!="setosa"),c(1:3,5)]
test3 <- iris[which(iris$Species!="versicolor"),]
test4 <- iris[which(iris$Species!="setosa"),]
test5 <- irisNote that entities may be species, genera, populations etc.
To plot the decision boundary using hyperoverlap_plot, the data cannot exceed three dimensions. For high-dimensional visualisation, see hyperoverlap_lda.
library(hyperoverlap)
setosa_virginica3d <- hyperoverlap_detect(test1[,1:3], test1$Species)
versicolor_virginica3d <- hyperoverlap_detect(test2[,1:3], test2$Species)To examine the result:
setosa_virginica3d@result #gives us the result: overlap or non-overlap?
#> [1] "non-overlap"
versicolor_virginica3d@result
#> [1] "overlap"
setosa_virginica3d@shape #for the non-overlapping pair, was the decision boundary linear or curvilinear?
#> [1] "linear"
hyperoverlap_plot(setosa_virginica3d) #plot the data and the decision boundary in 3dhyperoverlap_plot(versicolor_virginica3d) Note the points on the ‘wrong side’ of the boundary when comparing versicolor and virginica
To visualise overlap in n-dimensions, we need to use ordination techniques. The function hyperoverlap_lda uses a combination of linear discriminant analysis (LDA) and principal components analysis (PCA) to choose the best two (or three) axes for visualisation. To plot these using other methods (e.g. ggplot2), the point coordinates are returned as output, here named transformed_data.
setosa_virginica4d <- hyperoverlap_detect(test3[,1:4], test3$Species)
versicolor_virginica4d <- hyperoverlap_detect(test4[,1:4], test4$Species)To examine the result:
setosa_virginica4d@result #gives us the result: overlap or non-overlap?
#> [1] "non-overlap"
versicolor_virginica4d@result
#> [1] "overlap"
setosa_virginica4d@shape #for the non-overlapping pair, was the decision boundary linear or curvilinear?
#> [1] "linear"
transformed_data <- hyperoverlap_lda(setosa_virginica4d) #plots the best two dimensions for visualising overlap
transformed_data <- hyperoverlap_lda(versicolor_virginica4d) In three dimensions:
rgl.close() #close previous device
transformed_data <- hyperoverlap_lda(setosa_virginica4d, visualise3d=TRUE) rgl.close() #close previous device
transformed_data <- hyperoverlap_lda(versicolor_virginica4d, visualise3d=TRUE) #plots the best three dimensions for visualising overlapWe might want to know which species overlap in certain variables from an entire genus. To do this, we can use hyperoverlap_set and visualise the results using hyperoverlap_pairs_plot
all_spp <- hyperoverlap_set(test5[,1:4],test5$Species)
all_spp_plot <- hyperoverlap_pairs_plot(all_spp)
all_spp_plot
#> Warning: Use of `x$result` is discouraged. Use `result` instead.