R.Luminescence

Board for users of the R.Luminescence package

[FFM] Feedback on plot wanted

Only issues concerning the agemodels should be discussed here!

Moderator: skreutzer

Re: [FFM] Feedback on plot wanted

Postby cburow » Wed Feb 12, 2014 3:44 pm

Thanks to your feedback I think the plot greatly improved. For the sake of completeness I'd like to share the latest version of the plot. It was quite a struggle with the axis-scalings of the first plot, but I think it now works reliably.

Image

Best

Christoph
User avatar
cburow
 
Posts: 42
Joined: Thu May 16, 2013 1:39 pm
Location: Cologne

Re: [FFM] Feedback on plot wanted

Postby skreutzer » Thu Feb 13, 2014 4:11 pm

....perfect....

Sebastian
--
Package Developer
skreutzer
 
Posts: 69
Joined: Tue May 14, 2013 4:19 pm

Re: [FFM] Feedback on plot wanted

Postby questio » Tue Mar 11, 2014 3:05 pm

Hi Christophe,

The evolution from your first plots to the last is impressive in that at each step your message gets easier to grasp and appreciate.
By your last graph, the heat-map of proportions is redundant and one can then directly compare the different distributions with the BIC and llik values directly and appreciate the converging from overfitting.
There is one (minor) point that I feel might be slightly misleading and that is the colour coding.
The different colours helps separate the peaks and to keep the colours consistent from one graph to the next is a helpful visual guide. However, attributing one colour to a component number has has no meaning or purpose.That would suggest that as peaks appear after each incremental increase in components, they remain distinct, but this is not always the case, and is not necessarily important for the argument.
At n=2, the red (20%) peak is the n=1 peak rather than the blue peak (80%) and one could argue that it should be the opposite. Inverting the codes would however have no bearings or consequences to what you are presenting. ;)

Thanks for sharing that evolutionary journey!
Danièle
questio
 
Posts: 29
Joined: Tue Mar 11, 2014 7:47 am

Re: [FFM] Feedback on plot wanted

Postby cburow » Fri Mar 14, 2014 12:53 pm

Dear Danièle,

many thanks for the kind words and your suggestions!

1.) Barplot: Proportion (%)
I agree that the barplot is redundant, especially so when the normal distributions are weighted by the components proportion. But then again, I think the barplot is still valuable because it provides numerical values on the proportion whereas in the first plot(s) it is not always easy to estimate the proportioning. As a compromise, I added the option to turn plotting of the barplot on or off. This way, the user has more flexibility and can decide on his own wether the plot provides any beneficial information or not.

2.) Color coding
Thanks for pointing out the implications of the color coding, which is something I haven't really thought about. The intention for color coding the components was indeed what you first said: a helpful visual guide. To avoid misintepretations and also being dissatisfied by the visual aesthetics of the colors, I added an option to use either the aforementioned colors, greyscales or no colors at all. If you go for the later, the normal distributions are only plotted as dashed lines.

For quick illustration:

Image

In the first two plots the normal distributions aren't weighted by their proportions, so it might be viable to keep the barplot. In the last plot no color coding is used and because the normal distributions are weighted by the corresponding proportions the barplot is removed.

Thanks for your helpful suggestions. Hopefully, the latest changes are in accordance with what you had in mind.

Best

Christoph


PS: There are still some minor annoyances, especially with regard to the placement of the legends. Hopefully, R will eventually listen to what I want from it... However, the flashy colors and inappropriate line widths in the plot above are just an artefact of uploading the picture to the picture hoster.
User avatar
cburow
 
Posts: 42
Joined: Thu May 16, 2013 1:39 pm
Location: Cologne

Re: [FFM] Feedback on plot wanted

Postby questio » Thu Mar 20, 2014 4:26 am

Hi Christoph,
Your plots are brilliant and my reservation about the colour coding was not that there should be no colours (I personally enjoy colours and your choice was really good) but more that they should not be linked to the number of components. You had a legend with colour per component number and that was the point I was not happy about for an otherwise brilliant visualisation.
As you pointed out, the bar chart is not redundant as the proportion is linked to the area of the peaks and not their heights so that you definitely need a cue to the actual proportion… Oops!
I thought that maybe if the y axis scaling for each window for each component was standard you would see the “decline” and “rise” of some peaks with respect to others. The blue peak declines from about 80% to 10% which is not that visible on the graph. But then, with a standard scale the peaks would get smaller and smaller and the picture might get pretty messy. So your option is the neatest, and the bars are necessary.
You manage to make R listen to you more than many of us, but more importantly, you know what to ask it to do!
Thanks!
Danièle
questio
 
Posts: 29
Joined: Tue Mar 11, 2014 7:47 am

Re: [FFM] Feedback on plot wanted

Postby questio » Thu Jul 21, 2016 8:44 am

Hi Christoph,

I have been using very happilythe plot that you devised but despite setting the pdf.colors="colors",or "none", the plot appears only in greyscale.
Have you disbaled the option or am I using the wrong argument?
The colour plot (with no colour legend) is more attractive than the greyscale. ;)

Thank you,

Danièle
questio
 
Posts: 29
Joined: Tue Mar 11, 2014 7:47 am

Re: [FFM] Feedback on plot wanted

Postby cburow » Fri Oct 07, 2016 3:47 pm

Hi Danièle,

sorry for the slow response, this totally slipped my attention... :oops:

I just checked and indeed there was a bug which basically resulted in that arguments meant to modify the plot were ignored. This has been fixed and will officially be available in the next major release (0.7.0) of the package.

As you are familiar with installing the developer version from GitHub you may also install the latest development version via:
Code: Select all
devtools::install_github("r-lum/luminescence@dev_0.7.0")


Again, sorry the late response!

Best

Christoph

----------
Change commit: https://github.com/R-Lum/Luminescence/c ... 82abcc84ac
User avatar
cburow
 
Posts: 42
Joined: Thu May 16, 2013 1:39 pm
Location: Cologne

Previous

Return to Agemodels

Who is online

Users browsing this forum: No registered users and 1 guest

cron