Optimistic Exploration in Reinforcement Learning Using Symbolic Model Estimates