Making machine learning more accessible has been an area of focus lately, with the emergence of online courses, comprehensive textbooks, and frameworks to simplify the process. However, machine learning remains a difficult problem. Implementing existing algorithms and models for new applications requires not only understanding the math but also choosing the right tools for the task. Debugging in machine learning is exponentially harder because there are multiple dimensions of potential bugs, including algorithm, implementation, data, and the model itself. Furthermore, debugging cycles can be long, delaying feedback on potential fixes. Developing an intuition for debugging is crucial in machine learning projects, and effective debugging is a key skill for success in this field.
https://ai.stanford.edu/~zayd/why-is-machine-learning-hard.html