Train best versions of each strategy

[marcharper]

Ideally we would train the strategies with the smallest numbers of parameters that perform well -- see bias-variance tradeoff for context.