Conversation

there are apparently bugs in one or more of (a) numpy (b) openblas (c) mkl (d) anaconda that make it so some of my students get errors from calculating cosine distance between word vectors (I think the norm() function is returning negative numbers for no good reason sometimes) and the process of figuring out the problem is very annoying. like it's a dot product and two norms? how can you mess up those two extremely simple functions