Efficient matrix representations for working with tabular data
copied from cf-staging / tabmattabmat provides efficient tools for representing tabular data as a mix of dense, sparse and categorical subcomponents, resulting in 100x faster performance on some machine learning problems!