![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif)
At the start I would probably just use a straight random generated, but I thought about statistically modelling and if some of my assumptions are true I might try a few tricks. For example, I felt compelled by thinking about odds to not choose the same thing twice, if I kept track of my win/loss I might be able to verify this is true to a useful degree. My statistics is not that great, but I believe that if the compound event of 3*3 (3 choices, 3 games) has more possibilities than the outcome of 3 + 2 + 1 for the reducing unique set. Even though I don't get to see the choices of the player beforehand, I could gain an edge by selecting non-unique sets. Kind of like the http://en.wikipedia.org/wiki/Monty_Hall_problem - This could be retarded though I haven't thought it through all the way yet.