data partition

[shengoy]

Hi !
How could I split the data for Train/dev/test?

[Hellisotherpeople]

Wow, I am sorry for not seeing this earlier!

On the off chance that you are still interested or care about this problem - I'd say that the most "fair" way to do it might be to randomly sample 10-20% from each of the year splits.

Possibly better, if you are someone from the competitive debate community, would be to randomly sample that same 10-20% but at the level of each individual file. This would prevent the random sampling from favoring certain files/annotators over others and would hopefully maximize the diversity of the samples.