from models.han.__main__ import run_main
from models.han.args import get_args
from utils.postprocessing import process_json_results
if __name__ == '__main__':
args = get_args()
if args.num_folds < 2:
raise ValueError("Number of folds must be greater than 1!",
args.num_folds)
orig_metrics_json = args.metrics_json
for fold in range(0, args.num_folds):
print('On fold', str(fold))
args.fold_num = fold
if orig_metrics_json:
args.metrics_json = orig_metrics_json + '_fold' + str(fold)
run_main(args)
# summarize fold results and save to file
process_json_results(orig_metrics_json, orig_metrics_json + '_summary.tsv',
'test')
args = get_args()
if args.num_folds < 2:
raise ValueError("Number of folds must be greater than 1!",
args.num_folds)
orig_metrics_json = args.metrics_json
for fold in range(0, args.num_folds):
print('On fold', str(fold))
num_train_restarts = 0
args.fold_num = fold
orig_seed = args.seed
if orig_metrics_json:
args.metrics_json = orig_metrics_json + '_fold' + str(fold)
training_converged = run_main(args)
while not training_converged and num_train_restarts < args.num_train_restarts:
num_train_restarts += 1
args.seed += 10
print('Rerunning fold', fold, 'with new seed', args.seed)
training_converged = run_main(args)
args.seed = orig_seed
# summarize fold results and save to file
process_json_results(orig_metrics_json,
orig_metrics_json + '_fine_summary.tsv',
'test',
label_suffix='_fine')
process_json_results(orig_metrics_json,
orig_metrics_json + '_coarse_summary.tsv',
'test',
label_suffix='_coarse')