Increase performance when calculating feature matrix?

Question

Does calculate_feature_matrix use any libraries such as numba to increase performance?

Answer 1

I am one of the maintainers of Featuretools. calculate_feature_matrix currently only uses functions from Pandas/Numpy/Scipy to increase performance over raw Python. There are several areas where using numba or Cython may help, particularly in the PandasBackend class and in individual feature computation functions.

However, doing so requires a C-compiler or compiled C code, and so adds extra complexity to the installation. Because of this complexity it's currently not high on our priority list, but we may consider adding it in the future.

Instead, we are more focused on scalability to larger datasets, which involves parallelization rather than subroutine optimization.