The number of sentences in the first file is $1110$, and the number of predictions in the second file is $1109$. So, there are $9$ extra predictions.
If we assume that each sentence has a unique prediction, then these predictions must come from the third file. Therefore, the third file must have $9$ more predictions than it should.
So, the number of predictions in the third file is $1109+9=\boxed{1118}$.
2条答案
按热度按时间h7appiyu1#
Hi @Mahhos :tsv文件的第一行是否有表头？如果有，那么你将得到1109个预测结果，不包括表头。
如果不是这种情况，请尝试在tsv文件中添加一个id列，并将其与预测结果进行匹配，以便找到缺失的句子。
nuypyhwy2#
你好@AbinayaM02和@Mahhos,你们找到这个问题的原因了吗？我们知道哪个预测缺失了吗？是第一行、中间还是测试集的最后一行？谢谢。
编辑：大家好，已经找到了问题。请查看：https://stackoverflow.com/questions/62161281/bert-skipping-the-1st-row-of-test-tsv-when-predicting