search

hil-se / Technical-Debt-Risks

0 stars 0 forks source link

First step #13

Open azhe825 opened 2 years ago

azhe825 commented 2 years ago
  1. replicate one sota approach -> F1
  2. repeat 1 on the corrected data
  3. easy + sota
    • performance for easy, F1, prec, recall, tp
    • performance for sota on hard SATDs, F1, prec, recall -> lower than the numbers reported
    • overall performance F1, recall, for easy + sota -> higher F1 than the numbers they reported -> paper
    • sota -> rank hard -> APFD (on hold)
ketakisbarde commented 2 years ago
  1. For the tosem paper, i found the paper's code (pushed to repo as CNN_SATD) and one other implementation (link below) https://github.com/goodchar/CNN-based_SATD
  2. Both of these are tensorflow implelemted and not keras which i used originally, so trying to compare code. Next, I will try to use ELMO embeddings and see if it improves F1.