qianc62 / Corsair

Counterfactual-inference-based Text-classification Debiasing Framework.
98 stars 4 forks source link

What's exactly the 5 initiation in 3.1 #6

Open sangyx opened 7 months ago

sangyx commented 7 months ago

Hi, thank you for sharing your code. I am curious about how the final result in Table 2 was calculated. In section 3.1 of the paper, it is mentioned that we report the average results over five different initiations in Table 2. Do you run experiments with five seeds or average results for five different $[\lambda_1, \lambda_2]$? Which epoch do you report?