That's funny, I went the opposite direction. I only use pandas if I need to group by and aggregate, or to use in plotting functions. When doing array operations, I pretty much always use numpy, figuring out how to broadcast and vectorize operations is one of the most fun parts of my job in my opinion.