DFT-DISPLAY:  Print diffusion trees, parameters, and hyperparameters.

Dft-display can print a representation of the diffusion tree(s)
underlying the training cases, the parameters for these diffusion
trees, and the overall hyperparameters for the model.

Usage:

  dft-display [ -b ] [ -h ] [ -p ] [ -l ] [ -t ] [ -n | -N ] [ -g[time] ]
              log-file [ index ]

If no index is specified, the last record in the log file is shown.
The -b option enables "bare" output for latent vectors, tree
structure, and non-terminal nodes, in which headers are omitted, some
values are printed to higher precision, and each line starts with the
number of the tree it applies to.  The other options enable display
of certain information, as follows:

   -h  Values of the overall hyperparameters
   -p  Parameters for each diffusion tree 
   -l  Latent vectors for each training case (if stored)
   -t  Structure of the diffusion trees underlying the training cases
   -n  Non-terminal nodes of the trees
   -N  Same as -n, but with node locations as well (if stored)
   -g  Graphical display of the tree structures (picture done with text)

More than one of these options can be specified (in any order, but
this does not affect the display order), though specifying more than
one is probably not useful if -b is specified.  The default if none of
these flags are specified is -h and -p.

Currently, the only overall hyperparameters are the noise standard
deviations for the variables (if they are real), along with the common
noise standard deviation controlling these.  The common noise standard
deviation is displayed first, followed by a colon and then the noise
standard deviations for each variable.

One set of parameters for a diffusion tree are the standard deviations
for each variable, along with the common standard deviation
controlling these.  These are displayed in the same format as noise
standard deviations.  The parameters of the divergence function are
also displayed, labled as c0, c1, and c2.

The latent vectors are displayed one per line if -l is specified,
preceded by the number of the case, from 1 (and the number of the
tree, if -b is specified).  If no data model was specified, the latent
vectors will equal the data vectors.  Otherwise, they will be the
values in terms of which the distribution of the observed data is
defined.  Note, however, that for real data, latent vectors may not be
present in the log file, depending on which Markov chain operations
were used (see dft-mc.doc).

The structure of the diffusion trees is displayed with -t in a
low-level format that is mostly useful for debugging, or for input to
another program (using -b).  The non-terminal nodes are displayed
first, numbered -1, -2, etc., followed by the terminal nodes
corresponding to training cases, numbered 1, 2, etc.  One line is
displayed for each node, which (after the tree number if -b is
specified) shows the node number, the number of the parent node (0 for
the root), and the divergence time for the node (1 for terminal
nodes).

The display of non-terminal nodes with -n or -N is also low-level,
showing (after the tree number if -b is specifed) the node number (-1,
-2, etc.), the indexes of the node's two children, the number of
points under the node (ie, the number of terminal descendents), and
the divergence time.  For -N, this is followed by the node location.

Pictures of the trees are displayed if -g is specified, in the format
illustrated below:

     5| 3| 2| 1
      |  |  | 4
      |  | 2
      |
      | 2| 3
      |  | 5

The numbers before "|" are the counts of observations in a sub-tree.
The "5" in the top-left of the above display indicates that there are
five observations altogether, the "3" and "2" in the next column are
the numbers of observations in the two sub-trees of the root, and so
forth.  The numbers at the ends of lines are not counts, but rather
are the indexes of the cases associated with the terminal nodes.

If a time in the interval (0,1] is given immediately after -g,
non-terminal nodes whose divergence times are later than this are
shown as just a list of indexes of cases that are in the sub-tree
headed by this node.  For instance, if -g.7 were specified, the tree
shown above might be displayed instead as follows:

     5| 1 4 2
      |
      | 2| 3
      |  | 5

Here, the first sub-tree of the root had a divergence time greater
than 0.7, and is therefore shown just as the list "1 4 2", of cases
within its sub-tree.  The divergence time of the second child is less
than or equal to 0.7, and so its children are shown separately.

The lines in these diagrams can become arbitrarily long; no attempt is
made to wrap them.

Similar output in Postscript is produced by the dft-dendrogram program
(see dft-dendrogram.doc).

            Copyright (c) 1995-2003 by Radford M. Neal