Grapth Matching using BMRM
--------------------------

Configuration options description:
1.  int    quadratic
    - Set the "1" if we want to use quadratic assignment, 0 otherwise.

2.  int    maxdata
    - The maximum number of training/test exampels we will allow. Set
      to "-1" to allow any amount of data (provided in dataset file).

3.  string Program.pairs
    - Whether to treat every possible pair as a training instance, or
      whether to consider alternating pairs. Valid values are
      {"ALTERNATING", "ALLPAIRS"}. In other words, if we have training 
      graphs 1,2,3,4 Then "ALTERNATING" will use the pairs (1,2), 
      (3,4). "ALLPAIRS" will use (1,2), (1,3), (1,4), (2,3), (2,4), (3,4). 

4.  int    nnodes 
    - The number of nodes inthe graphs being matched. The code assumes
      that both graphs have the same size. Valid values are nnodes > 1. 

5.  int    nfeatures
    - The dimensionalirt of the (unary) feature vector (there is
      either 0 or 1 quadratic feature in this code, depending on
      whether we are using linear/quadratic assignment). Valid values
      are nfeatures > 0.

6.  double delta
    - The convergence threshold for quadratic assignment with 
      bistochastic normalisation. Valid values are delta > 0 
      or -1 if we don't want to perform bistochastic normalisation.

7.  string Data.path
    - Path to the folder containing data i.e. images, adjacency
      matrices, etc. See the [ref 1] and its accompanied code for
      further details.

8.  string Data.list
    - Name of the file containing the list of examples. See the [ref 1] 
      and its accompanied code for further details. 


References:  
1. T. S. Caetano, J. J. McAuley, L. Cheng, Q. V. Le, A. J. Smola
   Learning Graph Matching
   IEEE Transactions on Pattern Analysis and Machine Intelligence,
   2009.
