OpenBLAS provides numpy with the ability to compute the matrix dot product as fast as Matlab does. In particular, OpenBLAS is a fork of the hand optimized gotoBLAS project. Simply clone it from here, and then run make. You will need to make sure you have a Fortran compiler - probably gfortran. You can then run make install PREFIX=/home/USER/.local to install the built libraries. Note that this also builds and installs LAPACK for you.

You may also want to build and install SuiteSparse/Metis for faster linear algebra. Matlab uses this for x = A\b and also SVD. Download SuiteSparse and Metis. Extract SuiteSparse and place Metis inside the SuiteSparse directory so that you have a structure of SuiteSparse/Metis-4.0.

You can then make SuiteSparse by running make. Unfortunately, SuiteSparse doesn't provide a utility to install the binaries in an easy manor, so you will need to run the following script: