corpus.py 文件源码

def load(corpus_csv_file: Path,
             sampled_training_example_count: Optional[int] = None) -> 'Corpus':
        import csv
        with corpus_csv_file.open(encoding='utf8') as opened_csv:
            reader = csv.reader(opened_csv, delimiter=',', quotechar='"', quoting=csv.QUOTE_MINIMAL)

            def to_absolute(audio_file_path: Path) -> Path:
                return audio_file_path if audio_file_path.is_absolute() else Path(
                    corpus_csv_file.parent) / audio_file_path

            examples = [
                (
                    LabeledExampleFromFile(
                        audio_file=to_absolute(Path(audio_file_path)), id=id, label=label,
                        positional_label=None if positional_label == "" else PositionalLabel.deserialize(
                            positional_label)), Phase[phase])
                for id, audio_file_path, label, phase, positional_label in reader]

            return Corpus(training_examples=[e for e, phase in examples if phase == Phase.training],
                          test_examples=[e for e, phase in examples if phase == Phase.test],
                          sampled_training_example_count=sampled_training_example_count)