Yuval - people have this evil connotation of what optimization means. It is certainly possible to put all of your 100 factors into one RS and tweak until you get the ultimate backtest result. This is, of course, one way of optimizing and is certainly prone to overfit. But what I suggest is not to tweak for the best backtest result, because that is simply memorization of past results and will likely fall on its face going forward. Instead, I suggest you perform a process where the objective is to minimize the number of factors, without significantly degrading the results with an eye to monotonic increase in rank per bucket. You keep adjusting the node weights, but some nodes will approach zero weighting. You take the opportunity to prune these nodes and start again. Typically you can end up with perhaps 25% of the original factors and still have the same or better backtest performance than when you started. Think of it as an exorcism, where memorization (excess factors) is the demon and the RS is the occupied body.