search

freesunshine0316 / nary-grn

Code regarding to our paper "N-ary Relation Extraction using Graph State LSTM"
96 stars 27 forks source link

Why is there such a big difference between the results of the test set and the dev set? #12

Open zll666 opened 5 years ago

zll666 commented 5 years ago

Hello, I use data file 1 as the test set, the rest for the train set, class_num=2, max_epoch=50, the dev result can be 88, but the test result only can be 73. Why are the results of the test set and the dev set so different? I think this is due to the different distribution of the test set and the dev set . In this situation, the test set results are not very good for testing the trained model.What do you think about and deal with this problem? Is there any problem with the division of Peng data itself?

freesunshine0316 commented 5 years ago

@zll666 I also observed large variances among different potions of Peng data. As part of your cross-validation, you have to create other "fof"s for testing other potions as well, before averaging the accuracies across all potions to obtain the final accuracy.

leslie991123 commented 4 months ago

FileNotFoundError： [Errno 2] No such file or directory： '/d/PycharmProject/nary-grn-master/peng_data/drug_gene_var/1/data_graph_1' 在添加数据集时出现这样的问题是什么原因呢