Skip to content

[R-gui] regression diagram builder thingie? Maybe like Doodle in Winbugs?

9 messages · Paul Johnson, Ian Fellows, Joe Cheng [RStudio] +4 more

Original
Paul Johnson
# 
I'd like to have an R program that allows users to design regression
equations by dragging variables around in a canvas, to interactively
build path diagrams that lead to regressions. The diagrams would
generate code that would run, and then results would somehow be linked
to the drawings, perhaps showing coefficients on the edges or such.

I'd like this thing to facilitate multilevel modeling if possible, but
I'd settle to just have a pallet of variable "nodes" that can be
pulled out of a box on the side and re-positioned in the canvas, with
arrows pointing in and out. If that could generate code to run glm,
I'd be happy.

Then I'd like to generalize this so that the nodes could represent
latent variables in a structural equation model.  Maybe I'd fiddle it
up to write code for fitting with lavaan's estimators.

Has anybody tried to do such a thing?  Anybody know where to start?

If you were doing this, which of the graphical programming
environments would you suggest?  I want this to work more-or-less well
on all platforms. I've Googled enough to know it is not easy to decide
which path to follow.  My first idea was to imitate the design of
programs that draw mind maps, but they are mostly based on Java, which
in my experience is hard to support for diverse platforms. Still, the
JGR project seems to do well with it, so I'm not absolutely opposed.

I've been looking at GTK2, QT4, WXwidgets and tcltk. Judging from what
I read in the email lists of various development projects, perhaps QT4
is the least troublesome multi-platform gui library, but it is not
entirely open/free.  QT, of course, is the underlying framework of the
KDE desktop.  My favorite editor, LyX, is written with QT, so I am
sure it works. The Gambit game theory project chose WX.  It appears to
me that tcltk is constantly on the brink of extinction, and yet new
versions pop out now and then.

In WinBUGS, there's a graphical model designer called Doodle that is
quite pleasing to me, but it seems there must be something wrong with
it because nobody boasts about it very much :(

pj
5 days later
Greg Snow
# 
Look at the TkIdentify function in the TeachingDemos package.  It does
not do what you want completely, but it shows an example of dragging
and dropping labels within a plot in R.  You could do something
similar with your variable names to drag and drop them to different
locations (and possibly arrows or lines).  Then you just need to
decide how to use the position information to build your model.
However, this seems rather complicated, how will you indicate
interactions or non-linearity?  I could see this for very basic
models, but not for more general ones.
On Tue, Jun 19, 2012 at 2:05 PM, Paul Johnson <pauljohn32 at gmail.com> wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Paul Johnson
    

    

      
      #
    

  


  

      
So, just to be blunt:

You like working with tcltk?

I mean, I know your package is written with it, but do like it? Do you
sometimes wish you were working with QT or GTK? That's the hard
decision for me at the moment.

pj

        
On Mon, Jun 25, 2012 at 12:19 PM, Greg Snow <538280 at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Duncan Murdoch
    

    

      
      #
    

  


  

      
On 25/06/2012 4:13 PM, Paul Johnson wrote:

            

      
      
      
    

        
One advantage of tcltk is that installation is easy:  R does it for you 
on Windows, and it's likely there on Linux.  Only MacOS needs a little 
user work, but not much.  RGtk2 has improved in the installation of 
support libs, but it is still a two-step process on Windows, and I would 
suspect it will fail for many users who don't have the right permissions.

Duncan Murdoch

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Adrian Waddell
    

    

      
      #
    

  


  

      
Hello all,

I can not speak for QT nor GTK, but for tcl/tk, Java2d and the HTML5 canvas I can.

The tk canvas widget offers you some nice things out of the box: i.e. you can draw some shape onto the tk canvas and give it a tag. You then can specify what should happen if a user moves the mouse over a shape with some specific tag, or clicks on it etc... (Event bindings).

In Java2d and the HTML5 canvas (to my knowledge), out of the box, you listen to events and then need to figure out yourself what shape/element the mouse was over at the time of the event. You probably would want to use a framework such as Piccolo2d for Java or another scene graph implementation, but the decision on the best framework might not be easy to make, and you also need to learn these frameworks. Often, these frameworks and also the Java2d library itself make use of patterns. Hence, if you are new to programming or have not read any books on patterns you might find the way things are done a bit complicated. In this case, tcl/tk is good.

You also need to consider what Ian previously wrote: tcl/tk works pretty much with most R installations.

Your project sounds quite involved. You may consider to use tk to create a working prototype and then assess the situation new.

Personally, I like tcl/tk, but I would choose something else for a large scale project. However then I would probably also stick to the language with the best integration for a particular toolkit, i.e. Java or C++ often times. You will find many threads on stackoverflow on toolkit1 vs toolkit2. Qt and Gtk+ are probably the most popular ones at this time. (And of course the HTML 5 canvas, which is early wave however).

If you decide on a toolkit, let me/us know and maybe I/we can give you some input on how to start.

Adrian

        
On 12-06-25 04:13 PM, Paul Johnson wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Gabor Grothendieck
    

    

      
      #
    

  


  

      
On Mon, Jun 25, 2012 at 4:20 PM, Duncan Murdoch

        
<murdoch.duncan at gmail.com> wrote:

            

      
      
      
    

        
One way that this could be improved would be if tcl/tk and the tcltk R
package were included in all R distributions or at least the ones that
the R core group has control over.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Greg Snow
    

    

      
      #
    

  


  

      
I do prefer to work in basic tcltk.  Part of that may be my
background.  I started doing Tk guis using Perl/Tk before using it in
R (and still sometimes refer to my Perl projects/books even when using
it in R).  I have looked at some of the other tools occasionally but
have yet to see any advantages in them that would make it worth my
time to learn another toolset (not saying that they are bad or have
nothing to add, just that the apparent effort for me to learn them is
more than the perceived benefit that I would get).  Now if I were
starting from scratch (not already having experience with Tk) then it
might make more sense to start with a different tool.

        
On Mon, Jun 25, 2012 at 2:13 PM, Paul Johnson <pauljohn32 at gmail.com> wrote: