Why does this correlation matrix not rearrange (corrr)?

Question

Using corrr to produce a Pearson correlation matrix, I get a nice data frame and can rearrange to have a organised-looking matrix. However, when I plot this with rplot, the rearrangement seems to be thrown out.

Here is a subset of the correlation data frame, with the matrix run:

data <- select(data,c(npqmax,npq_end,npq_slope_up,pi,npqmax,fvfm,phipsii_end))

> data
# A tibble: 861 x 6
   npqmax npq_end npq_slope_up    pi  fvfm phipsii_end
    <dbl>   <dbl>        <dbl> <dbl> <dbl>       <dbl>
 1   2.60   0.866         1.25 0.805 0.745       0.492
 2   2.92   1.02          1.27 0.801 0.753       0.485
 3   2.95   0.881         1.33 0.832 0.752       0.518
 4   2.56   0.846         1.34 0.811 0.736       0.488
 5   2.68   0.822         1.52 0.820 0.738       0.499
 6   2.58   0.876         1.32 0.809 0.740       0.486
 7   2.82   0.908         1.14 0.824 0.749       0.505
 8   2.93   0.997         1.29 0.803 0.749       0.476
 9   2.71   0.936         1.51 0.819 0.740       0.490
10   2.80   0.844         1.40 0.837 0.754       0.527
# ... with 851 more rows

### next run Pearson correlation

cormat <- correlate(data)

> cormat
# A tibble: 6 x 7
  rowname       npqmax npq_end npq_slope_up      pi    fvfm phipsii_end
  <chr>          <dbl>   <dbl>        <dbl>   <dbl>   <dbl>       <dbl>
1 npqmax       NA       0.240        0.0103  0.0820  0.249       0.0582
2 npq_end       0.240  NA            0.193  -0.716  -0.0492     -0.729 
3 npq_slope_up  0.0103  0.193       NA      -0.167  -0.293      -0.261 
4 pi            0.0820 -0.716       -0.167  NA       0.383       0.918 
5 fvfm          0.249  -0.0492      -0.293   0.383  NA           0.614 
6 phipsii_end   0.0582 -0.729       -0.261   0.918   0.614      NA   

### make a nice rearrangement 

cormat2 %>%
  rearrange(method = "MDS", absolute = FALSE) %>% 
  shave()

> cormat2
# A tibble: 6 x 7
  rowname      npq_end npq_slope_up  npqmax   fvfm     pi phipsii_end
  <chr>          <dbl>        <dbl>   <dbl>  <dbl>  <dbl>       <dbl>
1 npq_end      NA           NA      NA      NA     NA              NA
2 npq_slope_up  0.193       NA      NA      NA     NA              NA
3 npqmax        0.240        0.0103 NA      NA     NA              NA
4 fvfm         -0.0492      -0.293   0.249  NA     NA              NA
5 pi           -0.716       -0.167   0.0820  0.383 NA              NA
6 phipsii_end  -0.729       -0.261   0.0582  0.614  0.918          NA

Now I would plot this with rplot(shape = 15, colors = c("red", "green")) but instead of getting a plot like what's found on the corrr creator's blog:

()

I get something that looks not-so-arranged:

[

Any idea what's going wrong?

Thanks.

Answer 1

I used mtcars which is also the example in corrr's blog, and I get the same results:

library(corrr)
library(dplyr)
library(ggplot2)

cormat <- correlate(mtcars)
cormat2 <- cormat %>%
rearrange(method = "MDS", absolute = FALSE) %>% 
shave()

cormat2 %>% rplot(shape = 15, colors = c("red", "green"))

If you look at your matrix data.frame, the values that are NA are appearing now, which means the rows orders are messed up. Should be reflected to the author, below I make a few alterations to corrr:::rplot.cor_df :

newplot = function (rdf, legend = TRUE, shape = 16, colours = c("indianred2", 
    "white", "skyblue1"), print_cor = FALSE, colors) 
{
    if (!missing(colors)) 
        colours <- colors
    row_order <- rdf$rowname
    pd <- stretch(rdf, na.rm = TRUE)
    pd$x <- factor(pd$x,levels=row_order)
    pd$y <- factor(pd$y,levels=rev(row_order))
    pd$size = abs(pd$r)
    pd$label = fashion(pd$r)
    plot_ <- list(geom_point(shape = shape), if (print_cor) geom_text(color = "black", 
        size = 3, show.legend = FALSE), scale_colour_gradientn(limits = c(-1, 
        1), colors = colours), theme_classic(), labs(x = "", 
        y = ""), guides(size = "none", alpha = "none"), if (legend) labs(colour = NULL), 
        if (!legend) theme(legend.position = "none"))
    ggplot(pd, aes_string(x = "x", y = "y", color = "r", size = "size", 
        alpha = "size", label = "label")) + plot_
}

newplot(cormat2,shape=15,colours=c("#29c7ac","#c02739"))

Quick explanation, in the above function, there is a line stretch(rdf, na.rm = TRUE) where the correlation data.frame is melted but the order of your variables are not retained. I just added two lines to refactor them, there are other ways, but for your purpose, this should be ok.