Using reinforcement learning for streaming data
As discussed throughout earlier chapters, the challenge of building models on streaming data is to find models that are able to learn incrementally and that are able to adapt in the case of model drift or data drift.
Reinforcement learning is a potential candidate that could respond well to those two challenges. After all, reinforcement learning has a feedback loop that allows it to change policy when many mistakes are made. It is therefore able to adapt itself in the event of changes.
Reinforcement learning can be seen as a subcase of online learning. At the same time, the second specificity of reinforcement learning is its focus on learning actions, whereas regular online models are focused on making accurate predictions.
The split between the two fields is present in practice in the types of use cases and domains of application, but many streaming use cases have the potential to benefit from reinforcement learning and it is...
