Incomplete hypothesis set in banking77 dataset?
#2
by
aarabil
- opened
I noticed that the banking77 dataset contains 77 distinct labels, but each sample only includes 72 hypotheses. I was expecting to see all 77 (as is the case with the other datasets).
Could this be a bug or was it an intentional design choice?
Thanks!
aarabil
changed discussion title from
Incomplete Hypothesis Set in banking77 Dataset?
to Incomplete hypothesis set in banking77 dataset?
Hey, that's an intentional choice, I removed some labels because they were of low quality. You can probably find more info on this in the data preprocessing scripts here: https://github.com/MoritzLaurer/zeroshot-classifier
MoritzLaurer
changed discussion status to
closed