def main():
# Be careful not to commit the csv files to the rep
train_csv_path = "train.csv"
eval_csv_path = "eval.csv"
train_dataset = load_dataset('squad', split='train[0:499]')
eval_dataset = load_dataset('squad', split='validation[0:99]')
generate_csv(train_csv_path, train_dataset)
generate_csv(eval_csv_path, eval_dataset)
happy_qa = HappyQuestionAnswering(model_type="BERT", model_name="bert-base-uncased")
before_loss = happy_qa.eval(eval_csv_path)
happy_qa.train(train_csv_path)
after_loss = happy_qa.eval(eval_csv_path)
print("Before loss: ", before_loss.loss)
print("After loss: ", after_loss.loss)
def test_qa_test():
happy_qa = HappyQuestionAnswering()
results = happy_qa.test("../data/qa/test.csv")
assert results[0].answer == 'October 31st'
assert results[1].answer == 'November 23rd'
def test_qa_eval():
happy_qa = HappyQuestionAnswering(
model_type='DISTILBERT',
model_name='distilbert-base-cased-distilled-squad')
result = happy_qa.eval("../data/qa/train-eval.csv")
assert result.loss == approx(0.11738169193267822, 0.001)
def test_qa_train():
happy_qa = HappyQuestionAnswering(
model_type='DISTILBERT',
model_name='distilbert-base-cased-distilled-squad')
result = happy_qa.train("../data/qa/train-eval.csv")