Cloud Radio Access Networks (C-RAN) is an evolution in the base station architecture, mainly composed of two elements: The Base Band Unit (BBU) and the Remote Radio Head (RRH). The BBU is a centralized pool of computational resources to provide the signal processing and coordination functionality required by all cells, while the RRHs are light radio units that User Equipment (UE) connects to via the RAN. Many advantages are derived from this architecture, such as dynamic BBU-RRH associations and statistical multiplexing gains. In particular, the BBU-RRH association problem is crucial for reducing power consumption. In this paper, we focus on decentralized BBU-RRH association, which has not received attention in the literature. Therefore, the aim of this work is to propose a hybrid two-stage approach that includes a game theoretic framework for the BBU-RRH association, and a centralized scheme to set the adequate number of available BBUs. The game among RRHs is solved by two different algorithms. The first relies on the best response algorithm, namely H-BR-IACA. The second is based on a reinforcement learning method (the replicator dynamics), namely H-DR-IACA. We compare our devised solution to a centralized approach proposed in a previous work. The results of our proposition show close performance to the centralized method.