Number of times the model will cycle through the training data.

Number of training examples utilized in one iteration.

How much to change the model in response to the estimated error each time the model weights are updated.

Rate at which the learning rate decreases after each epoch.

Penalty term added to the loss function to encourage smaller weights and reduce overfitting.

Alpha parameter for the Leaky ReLU activation function, controlling the slope of the activation for negative inputs.

Fraction of input units to drop to prevent overfitting during training.

Number of epochs with no improvement after which training will be stopped.

Minimum change in the monitored quantity to qualify as an improvement.

Threshold to clip gradient norms to prevent exploding gradients during training.

Coefficient for the entropy term in the loss function to encourage exploration.

Level of noise added to the state inputs to improve robustness.

Reward value for the snake when it eats food.

Penalty value for the snake when it collides with itself.

Penalty value for the snake for each move it makes (encourages faster completion).

Penalty value for the snake if it fails to make progress towards the food.

Reward value for the snake based on its proximity to the food.

Size of the replay buffer that stores experiences for training.

Exponent for prioritizing experiences in the replay buffer.

Small value added to priorities to ensure all experiences have a non-zero probability of being selected.

Probability of selecting a random action instead of the best action during training.

Rate at which the exploration epsilon decreases after each episode.

Minimum value for the exploration epsilon to ensure some exploration is always present.

Weight of the curiosity-driven reward component in the total reward calculation.

Learning rate for the forward model, which predicts the next state given the current state and action.

Learning rate for the inverse model, which predicts the action taken given the current and next states.