This paper studies game-theoretic distributed transmit-power control in a cognitive radio network. It presents a hybrid-adaptive algorithm that interfaces Iterative Water-Filling with two learning algorithms: the Hedging Algorithm and the Historical Matching Algorithm. Iterative Water-Filling helps achieve a fast convergence whereas the learning algorithms help guard against exploitation. The learning algorithms employed are selected based on their performance in deterministic and probabilistic network environments. The hybrid-adaptive algorithm is shown to offer improvements on other methods published. It also performs better than Iterative Water-Filling and the learning algorithms taken in isolation. The main metric is the utility achieved by the players in the game-theoretic setting.