create_chemprot_bert.py does not reproduce the train.tsv, dev.tsv and test.tsv files

[yyzhuang1991]

My issue is that I could not reproduce train.tsv, dev.tsv and test.tsv of Chemprot in the bert_data.zip.

Let me take generating train.tsv as an example. I ran create_chemprot_bert.py to generate the file train.tsv with the following command line:

python blue/bert/create_chemprot_bert.py data/ChemProt/original data/ChemProt

(data/Chemprot/original is obtained from data_v0.2.zip at this link). However, the number of lines in the generated train.tsv is 19, 019, which is very different from the number of lines (19, 461) in the train.tsv in the bert_data.zip at the same link.

For example, "23580446.T4.T33" is found in my train.tsv file but not in the provided train.tsv file, and "23261590.T1.T21" is not found in my train.tsv but is found in the provided train.tsv. Notably, all cases that are mismatched in either file have the 'false' relation label.