Comparing different clustering methods

Aim of this document

This is a demonstration of the ggiraph package’s interactive visualisation capability. Also, since patchwork was recently put on CRAN, “side-by-side interactive plotting” just got easier!

I will perform the PCA, tSNE and UMAP dimensionality reduction on the iris data.

Just so you know that I know:

• Yes, iris data is overused, but hey! It is only a quick coding demo!

• Yes, I know about plotly, it also has the side-by-side option, but it is very painful when there are multiple plotting aesthetics.

library(ggiraph)
library(patchwork)
library(tidyverse)
library(Rtsne)
library(umap)

Computing the dimensionality reduction objects

## We:
## + convert the iris data to a tibble
## + remove one duplicated row
## + add unique identifier to each row

iris_tbl = iris %>% 
  as_tibble() %>% 
  distinct() %>% 
  tibble::rownames_to_column("id")

iris_tbl

## # A tibble: 149 x 6
##    id    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
##    <chr>        <dbl>       <dbl>        <dbl>       <dbl> <fct>  
##  1 1              5.1         3.5          1.4         0.2 setosa 
##  2 2              4.9         3            1.4         0.2 setosa 
##  3 3              4.7         3.2          1.3         0.2 setosa 
##  4 4              4.6         3.1          1.5         0.2 setosa 
##  5 5              5           3.6          1.4         0.2 setosa 
##  6 6              5.4         3.9          1.7         0.4 setosa 
##  7 7              4.6         3.4          1.4         0.3 setosa 
##  8 8              5           3.4          1.5         0.2 setosa 
##  9 9              4.4         2.9          1.4         0.2 setosa 
## 10 10             4.9         3.1          1.5         0.1 setosa 
## # … with 139 more rows

## Computing the PCA, TSNE and UMAP objects
iris_pca = prcomp(iris_tbl %>% select_if(is.numeric))
iris_tsne = Rtsne(iris_tbl %>% select_if(is.numeric))
iris_umap = umap(iris_tbl %>% select_if(is.numeric))

Preparing the plotting data and static plots

The main function we will use is geom_point_interactive from the ggiraph package. It behaves just like geom_point, but with two additional aesthetics. The data_id aesthetic is critical to link observations between plots and the tooltip aesthetic is optional but nice to have when mouse over a point. After making these plots, the girafe function using the same syntax in patchwork will allow us to make a pretty interactive plot!

## Getting two reduced dimensions for each method
iris_plotdf = iris_tbl %>% 
  mutate(
    pca_1 = iris_pca$x[,1],
    pca_2 = iris_pca$x[,2],
    tsne_1 = iris_tsne$Y[,1],
    tsne_2 = iris_tsne$Y[,2],
    umap_1 = iris_umap$layout[,1],
    umap_2 = iris_umap$layout[,2]
  )



p1 = iris_plotdf %>% 
  ggplot(aes(x = pca_1, y = pca_2, 
             colour = Species,
             tooltip = id,
             data_id = id)) +
  geom_point_interactive() +
  labs(title = "PCA")

p2 = iris_plotdf %>% 
  ggplot(aes(x = tsne_1, y = tsne_2, 
             colour = Species,
             tooltip = id,
             data_id = id)) +
  geom_point_interactive() +
  labs(title = "tSNE")

p3 = iris_plotdf %>% 
  ggplot(aes(x = umap_1, y = umap_2, 
             colour = Species,
             tooltip = id,
             data_id = id)) +
  geom_point_interactive() +
  labs(title = "UMAP")

p1 + p2 + p3

At this point, p1 + p2 + p3 are combined together as a static plot thanks to patchwork. But making it iterative is just one line.

The magical interactive visualisation

girafe(code = print(p1 + p2 + p3), 
       width_svg = 15, height_svg = 5)

Extensions

For the tidyverse fanatics (myself included), you will see that we have made three plots with similar structures and cringe a little bit. Here is another way to make similar plot by pivoting.

iris_long_plotdf = iris_plotdf %>% 
  tidyr::pivot_longer(cols = matches("1|2"), 
                      names_to = c("method", "dim"),
                      names_sep = "_") %>% 
  tidyr::pivot_wider(names_from = dim, 
                     values_from = value)

iris_long_plotdf

## # A tibble: 447 x 9
##    id    Sepal.Length Sepal.Width Petal.Length Petal.Width Species method   `1`
##    <chr>        <dbl>       <dbl>        <dbl>       <dbl> <fct>   <chr>  <dbl>
##  1 1              5.1         3.5          1.4         0.2 setosa  pca    -2.67
##  2 1              5.1         3.5          1.4         0.2 setosa  tsne   13.6 
##  3 1              5.1         3.5          1.4         0.2 setosa  umap   13.9 
##  4 2              4.9         3            1.4         0.2 setosa  pca    -2.70
##  5 2              4.9         3            1.4         0.2 setosa  tsne   16.1 
##  6 2              4.9         3            1.4         0.2 setosa  umap   12.0 
##  7 3              4.7         3.2          1.3         0.2 setosa  pca    -2.88
##  8 3              4.7         3.2          1.3         0.2 setosa  tsne   16.1 
##  9 3              4.7         3.2          1.3         0.2 setosa  umap   11.9 
## 10 4              4.6         3.1          1.5         0.2 setosa  pca    -2.74
## # … with 437 more rows, and 1 more variable: `2` <dbl>

p4 = iris_long_plotdf %>% 
  ggplot(aes(x = `1`, y = `2`, 
             colour = Species,
             tooltip = id,
             data_id = id)) +
  geom_point_interactive() +
  facet_grid(~method, scales = "free")

girafe(code = print(p4),
       width_svg = 10, height_svg = 5)

Session Info

sessionInfo()

## R version 3.6.1 (2019-07-05)
## Platform: x86_64-apple-darwin15.6.0 (64-bit)
## Running under: macOS Mojave 10.14.6
## 
## Matrix products: default
## BLAS:   /Library/Frameworks/R.framework/Versions/3.6/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/3.6/Resources/lib/libRlapack.dylib
## 
## locale:
## [1] en_AU.UTF-8/en_AU.UTF-8/en_AU.UTF-8/C/en_AU.UTF-8/en_AU.UTF-8
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] gdtools_0.2.1   umap_0.2.3.1    Rtsne_0.15      forcats_0.4.0  
##  [5] stringr_1.4.0   dplyr_0.8.3     purrr_0.3.3     readr_1.3.1    
##  [9] tidyr_1.0.0     tibble_2.1.3    ggplot2_3.2.1   tidyverse_1.3.0
## [13] patchwork_1.0.0 ggiraph_0.7.0  
## 
## loaded via a namespace (and not attached):
##  [1] Rcpp_1.0.3        lubridate_1.7.4   lattice_0.20-38   utf8_1.1.4       
##  [5] assertthat_0.2.1  zeallot_0.1.0     digest_0.6.23     RSpectra_0.15-0  
##  [9] plyr_1.8.4        R6_2.4.1          cellranger_1.1.0  backports_1.1.5  
## [13] reprex_0.3.0      evaluate_0.14     httr_1.4.1        pillar_1.4.2     
## [17] rlang_0.4.2       lazyeval_0.2.2    uuid_0.1-2        readxl_1.3.1     
## [21] rstudioapi_0.10   Matrix_1.2-18     reticulate_1.13   rmarkdown_1.18   
## [25] labeling_0.3      htmlwidgets_1.5.1 munsell_0.5.0     broom_0.5.2      
## [29] compiler_3.6.1    modelr_0.1.5      xfun_0.11         pkgconfig_2.0.3  
## [33] askpass_1.1       systemfonts_0.1.1 base64enc_0.1-3   htmltools_0.4.0  
## [37] openssl_1.4.1     tidyselect_0.2.5  fansi_0.4.0       crayon_1.3.4     
## [41] dbplyr_1.4.2      withr_2.1.2       grid_3.6.1        nlme_3.1-142     
## [45] jsonlite_1.6      gtable_0.3.0      lifecycle_0.1.0   DBI_1.0.0        
## [49] magrittr_1.5      scales_1.1.0      cli_1.1.0         stringi_1.4.3    
## [53] reshape2_1.4.3    farver_2.0.1      fs_1.3.1          xml2_1.2.2       
## [57] vctrs_0.2.0       generics_0.0.2    tools_3.6.1       glue_1.3.1       
## [61] hms_0.5.2         yaml_2.2.0        colorspace_1.4-1  rvest_0.3.5      
## [65] knitr_1.26        haven_2.2.0