RewardClippingConfig<T>
Options & Configuration · AiDotNet.Configuration
Configuration for reward clipping.
How It Works
For Beginners: Clipping rewards to a fixed range can stabilize training when reward magnitudes vary widely. The famous Atari DQN paper clipped rewards to [-1, 1].
Constructors
| Constructor | Summary |
|---|---|
RewardClippingConfig() |
Creates a new instance with default values. |
Properties
| Property | Summary |
|---|---|
T MaxReward { get; set; } |
Maximum reward value after clipping. |
T MinReward { get; set; } |
Minimum reward value after clipping. |
bool UseClipping { get; set; } |
Whether to clip rewards (vs just scaling them). |