Randomisations - swapping algorithm - very slow convergence for large data sets

[GoogleCodeExporter]

If a user specifies the rand_structured randomisation with a
richness_multiplier of 1 and richness_addition of 0 then the system will
attempt to match the existing richness patterns exactly. A consequence of
this is that it is extremely unlikely to be able to allocate labels across
groups that will exactly match the patterns.  This is where the swap
algortihm comes in, whereby labels are swapped in and out of groups until
the targets are met.  

However, the current swapping algorithm is inefficient and often diverges
instead of converging.  This is most noticable with large data sets (e.g.
with 20,000 groups and 600 labels).

The algorithm needs to be adjusted.  

Original issue reported on code.google.com by shawnlaffan on 7 Mar 2010 at 12:19

[GoogleCodeExporter]

Fix implemented in revision 1572.  Now swaps in a more targeted manner, such 
that
labels that can be allocated to an unfilled group are swapped first.

Original comment by shawnlaffan on 7 Mar 2010 at 12:20

• Changed state: Fixed
• Added labels: ****
• Removed labels: ****

[GoogleCodeExporter]

This also has flow-on effects on issue #76, as spatially structured 
randomisations
will have more need of swapping for strict richness targets.

Original comment by shawnlaffan on 7 Mar 2010 at 12:22

• Added labels: Performance
• Removed labels: ****