Ucb

ucb ¶

Upper confidence bound (UCB).

Classes¶

UCB ¶

UCB(
    regressor_distribution: BaseRegressorDistribution,
    kappa: float = 2.0,
    batch_size: int = 1,
    mc_size: int = 10000,
)

Bases: ConfidenceBound, BaseMaximum

The upper confidence bound (UCB).

This acquisition criterion is expressed as

M [x; κ] = E [Y (x)] + κ \times S [Y (x)]

where $Y$ is the random process modelling the uncertainty of the surrogate model $\hat{f}$ and $κ > 0$ .

Parameters:

regressor_distribution (BaseRegressorDistribution) –

The distribution of the regressor.
kappa (float, default: 2.0 ) –

A factor associated with the standard deviation to increase the mean value.
batch_size (int, default: 1 ) –

The number of points to be acquired in parallel; if 1, acquire points sequentially.
mc_size (int, default: 10000 ) –

The sample size to estimate the acquisition criterion in parallel.

Source code in src/gemseo_mlearning/active_learning/acquisition_criteria/maximum/_confidence_bound.py

def __init__(
    self,
    regressor_distribution: BaseRegressorDistribution,
    kappa: float = 2.0,
    batch_size: int = 1,
    mc_size: int = 10_000,
) -> None:
    """
    Args:
        regressor_distribution: The distribution of the regressor.
        kappa: A factor associated with the standard deviation
            to increase the mean value.
        batch_size: The number of points to be acquired in parallel;
            if `1`, acquire points sequentially.
        mc_size: The sample size to estimate the acquisition criterion in parallel.
    """  # noqa: D205 D212 D415
    self.__kappa = kappa
    super().__init__(regressor_distribution, batch_size=batch_size, mc_size=mc_size)

Attributes¶

MAXIMIZE `class-attribute` ¶

MAXIMIZE: bool = True

Whether this acquisition criterion must be maximized.

output_range `instance-attribute` ¶

output_range: float = max() - min()

The output range.

qoi `property` ¶

qoi: Any

The quantity of interest.

Functions¶

update ¶

update() -> None

Update the acquisition criterion.

Source code in src/gemseo_mlearning/active_learning/acquisition_criteria/maximum/base_maximum.py

def update(self) -> None:  # noqa: D102
    super().update()
    self._qoi = max(
        self._regressor_distribution.learning_set.output_dataset.to_numpy()
    )

Ucb