Approaching realism in social dilemmas: moody reinforcement learners can conditionally replicate human models