Datasets:
Scoring system
#2
by
eoggc
- opened
If the prompt only scores with integer values 1-2-3-4-5, how was it possible to keep exactly 10% top texts?
The model is regression model, so it classifier to arbitrary integer (altough during training it only saw values from 0-5).
To get the top 10% of the texts we simply took all documents for langauge, sorted per the score and took top 10%
hynky
changed discussion status to
closed