What graph is best suited for this data? (and how can I produce it in R)

Question

I have this df object (9x19),

clusters pearson..s pearson..c pearson..a spearman..s spearman..c spearman..a cosine..s cosine..c cosine..a    Z0e..s    Z0e..c    Z0e..a    Z1e..s    Z1e..c    Z1e..a    Z2e..s    Z2e..c    Z2e..a
       2  0.5220314  0.9619119  0.8902166   0.9619119   0.8566094   0.8566094 0.9619119 0.9253174 0.9619119 0.5177371 0.5177371 0.5177371 0.5177371 0.5177371 0.5177371 0.5177371 0.5371546 0.5177371
       3  0.9477222  0.9058999  0.8902166   0.9477222   0.9238237   0.9238237 0.9477222 0.9118745 0.8416729 0.5218447 0.5218447 0.5218447 0.5218447 0.5218447 0.5218447 0.5263256 0.5420090 0.5311800
       4  0.9342793  0.9029126  0.8909634   0.9223301   0.8035848   0.9118745 0.9058999 0.7735250 0.8274832 0.5261389 0.5412621 0.5412621 0.5308066 0.5308066 0.5308066 0.5308066 0.5412621 0.5308066
       5  0.8409261  0.7768857  0.8050784   0.9217700   0.7615758   0.7916355 0.8200149 0.7399178 0.8140403 0.5354742 0.5405153 0.5405153 0.5354742 0.5405153 0.5405153 0.5354742 0.5405153 0.5405153
       6  0.8065721  0.7462659  0.7923824   0.9001120   0.6719567   0.7901419 0.8065721 0.7261016 0.7826736 0.5403286 0.5403286 0.5403286 0.5352875 0.5403286 0.5403286 0.5403286 0.5403286 0.5403286
       7  0.7938760  0.6700896  0.7804332   0.8792009   0.6506721   0.7610157 0.8020911 0.7092980 0.7481329 0.5401419 0.5401419 0.5401419 0.5401419 0.5401419 0.5401419 0.5401419 0.5401419 0.5401419
       8  0.7854742  0.6671023  0.7182599   0.8590366   0.6491785   0.7520538 0.7936893 0.7078043 0.7406647 0.5311800 0.6161314 0.5311800 0.5311800 0.5085885 0.5237117 0.5311800 0.5115758 0.5237117
       9  0.7809933  0.6577670  0.7197535   0.8491412   0.6071695   0.7356236 0.7949963 0.6779313 0.7279686 0.5229649 0.6030620 0.5408887 0.5229649 0.5029873 0.5130695 0.5229649 0.5037341 0.5162435
      10  0.7823002  0.6338686  0.6898805   0.8267364   0.6023152   0.7057506 0.7823002 0.6301344 0.7160194 0.5154966 0.5537715 0.5319268 0.5154966 0.4985063 0.5067214 0.5123226 0.4992532 0.5067214

which shows the adjusted Rand index between a hierarchical clustering for a given number of clusters and the true groups in the original data. For example, df[1, 1]=0.5220314, is the adjusted Rand index for "Pearson's correlation" proximity measure along with "s" which is the "single" linkage method. Similarly, c is complete and a is average. "Spearman" is "Spearman's correlation", "Cosine" is cosine, "Z0e" is "Euclidean" for original data, whereas "Z1" and "Z2" are standardized versions of the original data.

Any recommendations for graphing are welcomed. I wish to compare firstly, the different proximity measures, and secondly, for each proximity measure, to compare the linkage method for each, for the different cluster values.

Answer 1

My first try would be this:

R
library(data.table)
library(ggplot2)

dat <- fread('data.tsv')
dat <- melt(data=dat, id.var='clusters', value.name='corr')
dat[, clusters := as.character(clusters)]
dat[, method := sub('\\.\\..*', '', variable)]
dat[, linkage := sub('.*\\.\\.', '', variable)]
dat[, variable := NULL]
dat
     clusters      corr  method linkage
  1:        2 0.5220314 pearson       s
  2:        3 0.9477222 pearson       s
  3:        4 0.9342793 pearson       s
  4:        5 0.8409261 pearson       s
  5:        6 0.8065721 pearson       s
 ---                                   
158:        6 0.5403286     Z2e       a
159:        7 0.5401419     Z2e       a
160:        8 0.5237117     Z2e       a
161:        9 0.5162435     Z2e       a
162:       10 0.5067214     Z2e       a

gg <- ggplot(dat=dat, aes(x=clusters, y=corr, colour=linkage, group=linkage)) +
    geom_point() +
    geom_line() +
    facet_wrap(~method)

In the ggplot call, you could try swapping clusters with method or linkage to see if it suits your needs better.

Also, it is rarely a good idea to have factors arranged in alphabetical order so I would re-order them in a more meaningful way.