No you shouldn’t.
But here is how I would be amazed if any of my colleagues knew
- How kernels in SVM work ?
- Which feature space transformation is best for a particular data set ?
- Why Naive Bayes is so Naive?
- Explain SVD in their sleep ?
- Know basic thing about cost functions and error margins .
- Build neural networks with feature grouping in dataset
The key is knowledge and not the tools. Libraries will evolve and they will be obsolete. But the fundamentals stay.
Originally Written on Quora.