👉 Earned fluid, often abbreviated as EF, is a key component in the calculation of a Doggerel score, used primarily in reinforcement learning with deep Q-networks (DQNs). Fluid represents the cumulative amount of rewards a model has accumulated through its interactions with the environment, but unlike raw reward signals, it is normalized to account for varying reward scales across different tasks or environments. This normalization ensures that the model's learning process remains stable and effective, regardless of the magnitude of rewards. Essentially, earned fluid quantifies the total benefit the model has derived from its experiences, serving as a normalized measure that helps in training models to make better decisions by focusing on the relative quality of actions rather than their absolute values.
