Figure 3 - uploaded by Josef Daniël Hamelink
Content may be subject to copyright.
Download
View publication
Physical computer memory (RAM) needed to support tabular reinforcement learning structures as a function of world dimensions and number of crates. These calculations are based on standard integers and floating point numbers in Python, both of which are stored as 8 bytes (64 bits).

Physical computer memory (RAM) needed to support tabular reinforcement learning structures as a function of world dimensions and number of crates. These calculations are based on standard integers and floating point numbers in Python, both of which are stored as 8 bytes (64 bits).

Source publication
Figure 2: Optimizing number of actions versus optimizing number of...
Figure 3: Physical computer memory (RAM) needed to support tabular...
Figure 4: Learning curves showing the cumulative reward per episode for...
Figure 5: Default seed, reward structure 1.
Figure 6: Learning curves for the prioritized sweeping agent on...
Testing the Limits of Tabular Reinforcement Learning with Bomberman
Article
Full-text available
  • May 2022
In this paper we show that reinforcement learning methods can be applied to challenging pathfinding problems that are conventionally solved by applying tree searching algorithms to them. We explore the strengths, weaknesses and limits of tabular reinforcement learning by using a Prioritized Sweeping agent to solve a bomberman problem. The main reas...
Cite
Download full-text

Contexts in source publication

Context 1
... amount of physical memory (RAM) needed to store Q and N tables is illustrated in figure 3. The Q table stores the agent's reward estimates for a given state and action. ...
Context 2
... the memory usage scales so exponentially the more crates we add, we decided to push the program as much as possible. So for each board size in figure 3, we ran our program for every crate count possible (given our time and memory limitations). to keep the results consistent, we fixed a seed for the environment, but not for the agent. ...