Reducing a Dataset on an Attribute

Hi there,

i wanna measure the influence of an certain attribute on a class. The problem is that this attribute has 2 instancetypes (A and and that A occures 80% and B occures 20 %.  I want to filter that Data set to get approx.  20 % and 20 % occurences of A and B (randomly choosen).

Does anyone knows a Operator-Combination that can reduce the Data set, only having 20 % A and 20 % B?

thanks in advance

user