My personal preference is to have out of sample data set be on the same symbol as the in sample data set, but with randomly chosen segments for each. That removes my concern for correlation between symbols, and places multiple instances of different market conditions into each set. Since the...