Implement new evaluation policies from literature

[heitor57]

For the time being, we have no literature evaluation policies for "interactive" recommendation systems in our framework, most of them are adaptations. We need to add more of these evaluation policies to ground our framework as a reliable benchmark approach for "interactive" recommendation systems.

• "Unbiased Offline Evaluation of Contextual-bandit-based News Article Recommendation Algorithms"
• "Accordion: A Trainable Simulator for Long-Term Interactive Systems"
• "Improving offline evaluation of contextual bandit algorithms via bootstrapping techniques"