"""
Active learning using ratio based confidence uncertainty measure
between classes in the posterior predictive distribution to
choose which observations to propose to the oracle
"""
from torch import Tensor
from pyrelational.informativeness import classification_ratio_confidence
from pyrelational.strategies.classification.abstract_classification_strategy import (
ClassificationStrategy,
)
[docs]
class RatioConfidenceStrategy(ClassificationStrategy):
"""Implements Ratio Confidence Strategy whereby unlabelled samples are scored and queried based on
the ratio confidence for classification scorer"""
[docs]
def scoring_function(self, predictions: Tensor) -> Tensor:
return classification_ratio_confidence(predictions)