Scalarization functions

Scalarization functions relying on numpy.

morl_baselines.common.scalarization.tchebicheff(tau: float, reward_dim: int)

Tchebicheff scalarization function.

This function requires a reference point. It is automatically adapted to the best value seen so far for each component of the reward.

Parameters:

• tau – Parameter to be sure the reference point is always dominating (automatically adapted).

• reward_dim – Dimension of the reward vector

Returns:

Callable – Tchebicheff scalarization function

morl_baselines.common.scalarization.weighted_sum(reward: ndarray, weights: ndarray) → float

Weighted sum scalarization (numpy dot product).

Parameters:

• reward – Reward vector

• weights – Weight vector

Returns:

float – Weighted sum