Score 9 out of 10
Read Joseph Bradbury's full review
It's going to handle most of what you can throw at it. The problem lies in the sparsity of some datasets. When you have over 5ish dimensions of sparse data, the size of the model gets bloated for not much value. In several cases, I have had to adapt my model to account for sparsity in a way that minimized use of space, but also made interacting with the data more cumbersome.