MH:
- Write JSS paper
- Write paper 3
- Once loon is on CRAN, do:
  + put 'loon' in Imports (in DESCRIPTION)
  + comment in the loon parts in NAMESPACE (see TODOs)
  + put all files from ./misc in the corresponding directories (./man, ./R)
  + comment in all parts in zenplot() (see TODOs in zenplot.R)
  + comment in all parts in extract.Rd (see TODO)
  + in zenplot.Rd:
    - mention 'loon' as a choice for argument 'pkg'
    - define ospace depending on 'loon' again
    - comment in all loon parts (see TODOs)

WO:
- check plots_loon.Rd, l_ispace_config.Rd, na_omit_loon.Rd if loon functions
  were described accurately
- Fix/adapt intro.Rmd
- Write paper 1
- Fix all *_*_loon functions (see grid/graphics) for the use of 'loc' etc.

Paper 1 (WO; JCGS):
        About the zenplots in general, motivation as in intro.Rmd (mention grid, loon)
Paper 2 (MH; Econometrics copula issue or Statistics & Risk Modeling):
        Application to detecting dependence as in dependence.Rmd
	=> submitted
Paper 3 (mostly MH; JSS):
        Describe the more specific features such as choosing your own turns

Paper 4 (jointly with Mu & Avinash; JASA):
        How to graphically determine independence between groups of variables
        - use pobs of within-group distances of pobs of original data
	- argue that n choose 2 many samples can also be useful if n is small
          (easier to detect dependence in small dimensions)
	- show numerical problems when margins of one group are 'qt(, df = 0.5)'
	- show that the problem can be solved with the method 'canberra' of
	  dist().
	- show that many distances are still equal if the original data has ties
          (rgeom()) and the distances are computed from the original data. This
	  is better if the pobs are applied first (distances then seem to have
	  less ties)
        - Consider two sectors; the first column in the second sector is
          dependent to each column of the first sector, but all columns in the
	  first sector are independent of all but the first column of the second
	  sector. How large can we choose the second sector so that we can still
	  visually determine/see the dependence between the sectors?
        - refer to Annals paper about distance correlation; does their test (see
          'energy' package) fail in cases we can easily spot visually?
	- maybe the theory behind can be dealt with by Christian and Bruno?
        - maybe also applicable to vine copulas or graphical models
Paper 5: A graphical atlas of copulas
         - Show a zen plot for each copula family. Each 2d plot shows a
           bivariate copula of that family with increasing dependence
	 - Different families could also form different groups of the same zen plot


### New ideas/features #########################################################

- write gridMerge which essentially builds gTree(children = gList(plot1call, plot2call))
- be able to write plot1d = list(fun = c("hist", "density", "label"),
                                 args = list(fun1 = , fun2 = , fun3 = ))
  and get overlaid 1d plots (same for 2d)


### List of features (determined from man pages) ###############################

1) extreme_pairs, extreme_pairs_graph:
- find so-many pairs with largest/smallest value in a symmetric matrix (and
  graph them (not too useful))

2) occupancy_to_human:
- convert an occupancy matrix to a more human-readable one

3) 1d plots: rug, points, jitter, density, boxplot, hist, label, arrow, rect, lines
   2d plots: points, density, axes, label, arrow, rect
   ... for packages 'graphics', 'grid', 'loon'

4) scale01:
- scaling data to [0,1] with "columnwise", "all", "pobs" methods

5) to_list, zenpath:
- to_list returns a list of (grouped) matrices which can be passed to zenplot
- zenpath returns a sequence of variables which can be used to index the data
  (via to_list()) for plotting via zenplot.

6) unfold, zenplot:
- unfold returns a list consisting of the path and details about the layout
- display layout via plot1d = "layout", plot2d = "layout"
- zenplot:
  + indicates groups (via lists of lists)
  + correctly deals with (partially and fully) missing data
  + sophisticated labels
  + your own 1d/2d functions
  + your own turns
  + packages graphics, grid, loon
  + colors
  + n2dcol choices
  + different zigzagging methods
  + first1d, last1d
  + width1d, width2d


### Further TODO ###############################################################

- Should we introduce 'height2d'?
- Font size issues (solution? probably requires to write a helper function in the number
  of 2d plots per row and column and set cex (or gpar(cex)) accordingly...)
- How do pobs look like when computed from a multiv. normal mixture?
- Cite papers in all help files!
- Maybe we can speed-up the computation of a full Eulerian and (thus?) avoid the
  dependence on PairViz?
- Apply in the context of M. Wainwright
- The dependence vignette shows that there is a 'star-shaped' connection of
  variable 296 with many others (especially weak tail dependence). Can zen
  plots deal with that? Would be cool if a turn was only made when it's a
  different variable and the layout would simply extend straight in the same
  direction (as a pairs plot) if it's the same variable.
  => A topic for a PhD student? design layout...