DFT-QUANTITIES:  Quantities from log files relating to diffusion trees.

The quantities below relating to diffustion tree models can be
obtained from log files (eg, for use in dft-plt).  The generic
quantities documented in quantities.doc are also available.  The
Markov chain quantities documented in mc-quantities.doc are defined as
well, but at present most of them are not meaningful for diffusion
tree models, since the standard Markov chain operations are not
supported.

The quantities specific to diffusion tree models are listed below.  If
"n" is present after the letter, it represents a numeric modifier,
optional with default 1 when in square brackets.

    tn   Array of n'th target values for training cases (numbered from 1).
         NOTE:  One cannot omit n, since if it's missing, it's the 't' 
         quantity described in quantities.doc.

    o[n] Array giving for each training case (numbered from 1) the latent
         value associated with variable n (numbered from 1).  Available
         only if latent vectors are stored.

    V[n] As an array, the variances of the n'th diffusion tree (from 1)
         for each variable (numbered from 1).  As a scalar, the top-level 
         variance hyperparameter that controls these variances.  

    v[n] Same as for V, but expressed in terms of standard deviations
         rather than variances.

    c[n] Array giving the parameters for the divergence function for the
         n'th diffusion tree (from 1).  At present, only parameters c@0, 
         c@1, and c@2 exist.

    N    As an array, the noise variances for each target variable 
         (numbered from 1).  As a scalar, the top-level variance 
         hyperparameter that controls these variances.  Valid only when 
         the targets are real-valued.  

    n    Same as for N, but expressed in terms of standard deviations
         rather than variances.

    dn   As an array, the divergence times for internal nodes of the n'th 
         diffusion tree (numbered from 1), sorted by increasing divergence 
         time.  The lower bound given for portion of the array desired is 
         ignored; a lower bound of 1 is always used.  For example, "d1@1:10", 
         "d1@:10", "d1@10", and "d1@5:10" all give the ten earliest divergence 
         times for tree 1.  As a scalar, the average divergence time over all
         internal nodes.  NOTE:  The tree number, n, cannot be omitted, since
         if it's missing, it's the 'd' quantity described in mc-quantities.doc.

    Dn   As an array, the depths for the training cases (numbered from 1)
         in tree n.  The depth is the number of links to successive parents 
         that need to be followed to reach the root.  (The depth will be zero 
         if there is only one training case, and otherwise will be at least 
         one, and no greater than the number of training cases minus one.)  '
         As a scalar, the average depth over training cases.  NOTE:  The 
         tree number, n, cannot be omitted, since if it's missing, it's the 
         'D' quantity described in mc-quantities.doc.

    a... Divergence time of last common ancestor of two training cases in
         a tree.  The numeric modifier following "a" is interpreted in
         terms of its decimal digits.  The first non-zero digit should be
         the number of the tree (from 1).  This should be followed by the
         numbers of the training cases (from 1), which must have the SAME
         number of digits.  For instance, "a10213" is the divergence time
         of the last common ancestor of training cases 2 and 13 in tree
         number 1.

    A... An indicator (0/1) of whether or not two training cases are more
         closely related in a tree than they are to a third training case.  
         The numeric modifier following "c" is interpreted in terms of its 
         decimal digits.  The first non-zero digit should be the number of 
         the tree (from 1).  This should be followed by the numbers of the 
         three training cases (from 1), which must have the SAME number of 
         digits.  For instance, "A2555008077" has the value 1 if in tree 
         number 2, training case 555 is more closely related to training 
         case 8 than either of them are to training case 77; otherwise, it
         has the value 0.  The three training cases must be distinct.

Note that the numbering of training cases starting from 1 is different
from some other modules in this software.  This is so these numbers
can be combined with zero for the root node and with negative numbers
for other non-terminal nodes.

            Copyright (c) 1995-2004 by Radford M. Neal