samialabed / rlcache

Cache Manager using Reinforcement Learning
9 stars 5 forks source link

Episodic design #1

Open samialabed opened 5 years ago

samialabed commented 5 years ago

Can model the system as a series of episodes.

This involves adding a step counter in each RL model and a scheduler.

The state has to be updated to include the number of observed states. The scheduler makes the decision of when to terminate an episode.

No changes needed to YCSB, simply an "episode" becomes just a checkpoint time. Need to investigate how often we need to do agent.update()