Ternary Interpolation / Smoothing

by Nicholas Hamilton 3/08/2015 | 8:12 15 Posted in Advanced, New Feature, User Request

For a long time, people have been sending me requests for a suitable smoothing / contouring / interpolation geometry be made available via ggtern, over and above the Kernel Density function. I am very pleased to say, that the recent Version 1.0.6 has this feature added. Let me demonstrate how it works.

This geometry & stat required an additional mapping, to ‘value’, in order to conduct the interpolation, which by default is done using multivariate linear regression as the mechanism for the interpolation:

#Start by setting up the environment
library(ggtern)
data(Feldspar)
theme_base <-  function(){
  list(
    theme_bw(),
    theme(legend.position      = c(0,1),
          legend.justification = c(0,1)))
}

#Demonstrate the simplest example
plot <- ggtern(Feldspar,aes(x=Or,y=An,z=Ab)) + 
  geom_interpolate_tern(aes(value=P.Gpa,color=..level..)) +
  geom_point() + 
  theme_base() + labs(color="P/Gpa Model")
plot

#Start by setting up the environment

library(ggtern)

data(Feldspar)

theme_base <- function(){

list(

theme_bw(),

theme(legend.position = c(0,1),

legend.justification = c(0,1)))

}

#Demonstrate the simplest example

plot <- ggtern(Feldspar,aes(x=Or,y=An,z=Ab)) +

geom_interpolate_tern(aes(value=P.Gpa,color=..level..)) +

geom_point() +

theme_base() + labs(color="P/Gpa Model")

plot

There are plenty of options to tailor how the interpolation gets done, lets reduce the spacing between the contours, and add some more vibrant colours:

#Demonstrate the use of some options
plot <- ggtern(Feldspar,aes(x=Or,y=An,z=Ab)) + 
  geom_interpolate_tern(aes(value= P.Gpa,fill=..level..),
                        colour   = "white",
                        formula  = value~poly(x,y,
                                              degree=2,
                                              raw=TRUE),
                        method   = "lm",
                        binwidth = 25,
                        buffer   = 1.5,
                        n        = 200) +
  geom_point() +
  theme_base() + 
  scale_fill_gradient(low="green",high="red") + 
  labs(fill="P/GPa Model")
plot

#Demonstrate the use of some options

plot <- ggtern(Feldspar,aes(x=Or,y=An,z=Ab)) +

geom_interpolate_tern(aes(value= P.Gpa,fill=..level..),

colour = "white",

formula = value~poly(x,y,

degree=2,

raw=TRUE),

method = "lm",

binwidth = 25,

buffer = 1.5,

n = 200) +

geom_point() +

theme_base() +

scale_fill_gradient(low="green",high="red") +

labs(fill="P/GPa Model")

plot

For further information, see the following:

?stat_interpolate_tern
?geom_interpotate_tern

1 2	?stat_interpolate_tern ?geom_interpotate_tern

Tags:Interpolation, smoothing

Nick Hamilton

Author of the ggtern package. PhD candidate at the University of New South Wales in the School of Materials Science and Engineering. My field is in the simulations of metallic glasses via several advanced simulation techniques such as QMD and RMC.

Comments

Mike Slattery 4/08/2015 | 3:24

Hello,
Just curious,
I can’t seem to find the function:
geom_interpolate_tern() anywhere. It doesn’t appear to be part of ggtern? And the queries
?stat_interpolate_tern
?geom_interpotate_tern

appear to go nowhere. Am I missing something?

thanks for very informative posts!!
best,
Mike

Reply

Nicholas Hamilton 4/08/2015 | 3:28

Have you upgraded to Version 1.0.6, it was only accepted by CRAN yesterday…

Reply

John Connolly 16/12/2015 | 4:04

I wish to supply a function of the three proportions p1, p2 and p3, calculate value of the function at the grid points for a simplex within ggtern and then display the function using a contour plot. I do not see how to compute the function values within ggtern. Could you please suggest a solution. Thanks

Reply

Nicholas Hamilton 18/12/2015 | 11:38

I would suggest applying the function to the desired points (mesh) outside of ggtern to produce a dataset of points and their corresponding values, and then supplying this dataset to the geom_interpolate_tern(…) geometry as documented above.

Reply

Abduljamiu Olalekan Amao 17/12/2015 | 5:03

Hello,

I have a requirement to use polygons to show the relationships in my data ( it’s a very small data ). I tried for a whole day using geom_polygon_tern to do this but wasn’t successful. Can you help please.

Reply

Nicholas Hamilton 18/12/2015 | 11:39

Can you show me what you have tried so far?

Reply

Abduljamiu Olalekan Amao 19/12/2015 | 7:39

Hi Nicholas,

Basically i was successful in plotting my data on the ternary diagram. What i was trying to add is a polygon to cover the area where the data point appear on the plot. I’ve been able to add this using ppt. but it will be nice if i can do everything directly on R. it sad i can’t attached the final plot so you can see what i’m trying to explain above but here is what i used
“## for tran.tern
#Algal.Mat Sheltered.Inner.Lagoon Foreshore Back.Pool Salicornia.Marsh Tidal.Flat Tidal.Channel
ggtern(data=tran.tern,aes(x=Hyaline,y=Porcelaneous,z=Agglutinated, color=Sub.Environments))+
geom_point(size=10)+
theme_bw()+
theme_hidetitles()”

Reply

Johannes 18/12/2015 | 8:20

Hello,

Is there a way to overplot the grid instead of having transparent contour filling?

Thanks you!

Reply

Nicholas Hamilton 18/12/2015 | 11:40

If I am understanding your question correctly, you are wishing to change the z-order of the grid, so that it runs over the top of any layers? Unfortunately, this is not possible at the moment.

Reply

komayafa 29/12/2015 | 9:21

Trying to install ggtern in R-3.2.3 (64-bit) returns the following error:

* installing *source* package ‘ggtern’ …
** package ‘ggtern’ successfully unpacked and MD5 sums checked
** R
** data
** preparing package for lazy loading
Warning: replacing previous import by ‘grid::arrow’ when loading ‘ggtern’
Warning: replacing previous import by ‘grid::unit’ when loading ‘ggtern’
Warning: replacing previous import by ‘scales::alpha’ when loading ‘ggtern’
Error in get(x, envir = ns, inherits = FALSE) :
object ‘coord_transform.cartesian’ not found
Error : unable to load R code in package ‘ggtern’
ERROR: lazy loading failed for package ‘ggtern’

Is there a fix in the works?

Reply

Admin 2/01/2016 | 10:13

This problem is from the new version of ggplot2, requiring significant changes to ggtern, which I am in the process of working on. Sorry about this, but until I post an update, you wil need to revert to the prior version of ggplot2 (ie 1.1 or whatever it is)

Reply

fabio 2/01/2016 | 10:09

After re-installing R, I can not install ggtern again :
Error in get(x, envir = ns, inherits = FALSE) : object ‘coord_transform.cartesian’ not found
Maybe an issue related to the new version (2.0) of ggplot?

Reply

Admin 2/01/2016 | 10:14

Fabio, totally, Im having to re-write a heap of stuff. Hoping to have this resolved soon. Appologies.

Reply

fabio 10/01/2016 | 10:16

Thanks for your work, and I hope you the best! Keep ggtern rockin’ because is a great piece of software!

Reply

Ahmed 2/10/2017 | 5:44

Why didn’t you publish this before I finish my PhD 🙂 it would have saved me a lot of time! Wonderful job. You don’t have to answer this question, I am just curious about the interpretation of the results? Is there any formulation recommendation given these parameters?

Reply