Commands/bbreak

From TNT

BBREAK
Version: 1.1.
Menu path: to do.

Bbreak
Perform branch-swapping, using pre-existing trees as starting point. 
   Use "bbreak=options;" (change settings and run) or "bbreak:options;" 
   (change settings, don't run).  This swaps according to current settings 
   of suboptimal, constraints, and collapsing. 
     
   Basic options are: 
     tbr         use TBR 
     spr         use SPR 
    [no]fillonly swap until tree-buffer is filled, and then stop. 
     

   Fine-tuning options are: 
    [no]safe     The "safe" option uses a slower (but safer) method for 
                 updating buffers when finding a better tree under TBR 
                 (default is "nosafe"). 
    [no]skipspr skips the SPR phase on a single tree when doing multiple 
                 RAS+TBR saving several trees per replication; "skipspr" 
                 is useful in conjunction with "nosafe", but not so much 
                 with "safe" because then the initial portion of TBR 
                 (when better trees are being found often) gets slowed down 
                 and SPR doesn't.  The "nosafe" option only makes a  
                 difference for very large data sets; note that "skipspr" 
                 modifies the behaviour of both "mult" and "xmult". 
    [no]int N    There are two options, "int 1" and "int 2" (both options 
                 identify most relevant characters for a series of swaps, 
                 option 2 also reorders characters to try to save time). 
                 These options useful only for large matrices (>10,000) with 
                 large numbers of characters; otherwise they tend to produce 
                 slower swapping. 
    [no]preproc  with "preproc" the program tries to identify and effect 
                  first the clippings that would improve the tree the most; 
                 this has effect only on "mult" searches with hold=1 (and 
                 only when skipping the SPR phase), and TBR swapping from  
                 existing trees with mulpars off.  This may save a little  
                 time in the initial stages of the search for very large 
                 data sets, although the end gain is small. 
    clusters N   use node-clusters of N nodes.  As the data set becomes  
                 larger, clusters of more nodes produce faster TBR-swapping. 
                 When using clusters also for wagner trees, the same size as 
                 defined here is used.

---

Example: Perform branch-swapping with the trees in memory, using TBR. Using skipspr you will get a faster search but less intensive.