Source code for scorers.pipeline_scorer

import torch
from transformers import pipeline

from .positive_scorer import PositiveScorer


class PipelineScorer(PositiveScorer):
    """
    Feature class relying on the pipelines from Huggingface's transformers
    """

    def __init__(self, label, params, temperature=1.0):
        """initializes a PipelineFeature's instance

        Parameters
        ----------
        label: string
            expected positive label from the pipeline
        """

        self.label = label
        self.pipeline = pipeline(**params)
        self.temperature = temperature

    def log_score(self, samples, _):
        """computes the log-scores of the samples
        from the label returned by the pipeline

        Parameters
        ----------
        samples : list(Sample)
            list of samples to log-score

        Returns
        -------
        tensor of log-probabilities"""

        return torch.log(
                torch.tensor(
                    [[r_i["score"] for r_i in r if self.label == r_i["label"]][0]
                    for r in self.pipeline([s.text for s in samples], return_all_scores=True)]
                ).float()
            ) / self.temperature