In Co-clustering, there could be many local optimal where the algorithm may get struck resulting in sub-optimum results. Hence we applied a strategy called XEM strategy to run the EM algorithm. The various steps are defined as follows:
Do several runs of: "initialization followed by
short run of algorithm (few iterations/high tolerance)". This parameter is
named as "nbxem" in coclusterStrategy
function. Default value
is 5. We call this step as xem step.
Select the best result of step 1 and make long run of Algorithm(high iterations/low tolerance).We call this step as XEM step.
Repeat step 1 and 2 several times and select the best result.
The number of repetitions can be modified via parameter "nbtry" of
coclusterStrategy
function. Default value is 2.