The code simultaneously generates the results for all the stop-criteria presented in the paper.
The results are dumped to several files in a log folder.
We can check the results by reading the dump files after running the codes.
Note that the stop-point of the BUS-stop can be checked from the conf-sim and class-sim performances recorded in the log. 

The results of Table 2 can be generated by running the 'run_main_at_seed.py'
  Eg) python run_main_at_seed.py --gpu 0 --data SST-2 --k 50 --valid_k 25 --epochs 30 

The result of Figure 3 can be generated by running the 'run_kmain_at_seed.py'
  Eg) python run_main_at_seed.py --gpu 0 --data IMDB --k 50 --valid 0.25 

The results of Table 4 and 5 can be generated by running the 'run_kmain_at_seed.py'
  Eg) python run_skewed.py --gpu 0 --data SST-2 --K 100 --trn_neg 0.4 --tst_neg 0.6 
  (In this code, run_skewed.py, the 'K' means the number of all the labeled samples, not per class.)

** We plan to upload the easy version of this code after publication **