A Case Study with CICIDS2017 on the Robustness of Machine Learning against Adversarial Attacks in Intrusion Detection