def main(args):
if args.do_eda:
from theta.modeling import show_glue_datainfo
show_glue_datainfo(glue_labels, train_data_generator, args.train_file,
test_data_generator, args.test_file)
else:
trainer = AppTrainer(args, glue_labels)
# --------------- Train ---------------
if args.do_train:
train_examples = load_train_examples(args.train_file)
eval_examples = load_eval_examples(args.eval_file)
trainer.train(args, train_examples, eval_examples)
# --------------- Evaluate ---------------
elif args.do_eval:
eval_examples = load_eval_examples(args.eval_file)
model = load_model(args)
trainer.evaluate(args, model, eval_examples)
# --------------- Predict ---------------
elif args.do_predict:
test_examples = load_test_examples(args)
model = load_model(args)
trainer.predict(args, model, test_examples)
save_predict_results(args, trainer.pred_results,
f"./{args.dataset_name}_predict.json",
test_examples)
def do_predict(args):
args.model_path = args.best_model_path
test_examples = load_test_examples(args)
model = load_model(args)
trainer.predict(args, model, test_examples)
reviews_file = save_glue_preds(args, trainer.pred_results,
test_examples)
return reviews_file
def do_eval(args):
args.model_path = args.best_model_path
_, eval_examples = load_train_val_examples(args)
model = load_model(args)
trainer.evaluate(args, model, eval_examples)