Method parse_result

python/LAC/models.py:272–303

将SEG模型输出的Tensor转为明文

(self, lines, crf_decode, dataset, words_length)

        return tensor, words_length
    
    def parse_result(self, lines, crf_decode, dataset, words_length):
        """将SEG模型输出的Tensor转为明文"""
        offset_list = crf_decode.lod[0]
        crf_decode = crf_decode.data.int64_data()
        batch_size = len(offset_list) - 1

        batch_out = []
        for sent_index in range(batch_size):
            begin, end = offset_list[sent_index], offset_list[sent_index + 1]

            sent = lines[sent_index]
            tags = [dataset.id2label_dict[str(id)]
                    for id in crf_decode[begin:end]]
            tags_for_rank = []

            if self.custom:
                self.custom.parse_customization(sent, tags)

            sent_out, tags_out = [], []
            for ind, tag in enumerate(tags):
                # for the first char
                if len(sent_out) == 0 or tag.endswith("B") or tag.endswith("S"):
                    sent_out.append(sent[ind])
                    tags_out.append(tag[:-2])
                    continue
                sent_out[-1] += sent[ind]
                # 取最后一个tag作为标签	
                tags_out[-1] = tag[:-2]

            sent_out = [''] if len(sent_out) == 0 else sent_out
            batch_out.append([sent_out, tags_out, tags_for_rank])
        return batch_out

class RankModel(Model):
    """Docstring for Rank Model"""