Efficient resource allocation is critical for ensuring quality-of-service in wireless networks. This paper proposes a machine learning based approach for intelligent allocation of spectrum, power and antennas in wireless systems. We develop a recurrent neural network model that can learn complex patterns in wireless traffic and channel conditions. The model is trained using deep reinforcement learning to dynamically optimize resource allocation policies. Simulation results demonstrate superior performance of our approach compared to conventional heuristic algorithms. The intelligent resource manager significantly improves spectral efficiency, energy efficiency and latency.