Discriminator - number of articles as an input

[alantaitz]

I have noticed that if I try to classify a very small number of articles (a jsonl file of < 15 articles), the output will be scores of [0. 0.] always. In trying to classify an individual article (and getting always [0. 0.]) I ended up multiplying the line of the article by 30, and this way I would get a reasonable output. For example
[0.3 0.7]
[0.3 0.7]
[0.3 0.7]
[0.3 0.7]
...
Has anyone else seen this problem? Any ideas?

[billpwchan]

Same! I always ended up with [0. 0.] result using the provided jsonl file. Have you figured out how to solve this issue?

[alantaitz]

I haven't! I used it as it was copy pasting 30 times a single articles to obtain a result. I could verify that this way yielded an accurate result. Strange error!