Per #16, add back the class token to the ParticleTransformer code, with some more careful checks.
Also removed the outdated training and dataset classes.
jet regression (red is with class token), no significant difference:
decay mode (blue is with class token), no significant difference:
binary classification (red is with class token, still training), no significant difference:
So far it does not look like it has a major effect, but for the sake of being close to the original implementation, it might be useful to add it back still.
Per #16, add back the class token to the ParticleTransformer code, with some more careful checks. Also removed the outdated training and dataset classes.
jet regression (red is with class token), no significant difference:
decay mode (blue is with class token), no significant difference:
binary classification (red is with class token, still training), no significant difference:
So far it does not look like it has a major effect, but for the sake of being close to the original implementation, it might be useful to add it back still.