Learning Against Non-Stationary Opponents in Double Auctions