Education on machine learning and data science has drawn a lot of attention in both higher education and vocational training. Although various tools and services such as Jupyter Notebook and Google Cloud's AI have been developed for building and training models, they are not suitable for direct use in educational settings. For example, teachers expect a platform where they can easily distribute and grade programming assignments, and students want to quickly start coding and training models without the burden of setting up an environment. To this end, we develop MLadder, an online training system for machine learning and data science education. Specifically, we seamlessly integrate two open-source software, CodaLab and Jupyter Notebook, which are used for hosting assignments and building models, respectively. Moreover, we devise several methods to make the system lightweight and scalable, so that it can be deployed on-premises even with limited resources. We have used MLadder in the machine learning and data science courses in our school and facilitated both teaching and learning.