Python for Finance: Analyze Big Financial Data

There is a clear winner: the multithreaded numexpr implementation f6. Its speed

advantage, of course, depends on the number of cores available. The vectorized NumPy

version f4 is slower than f5. The pure Python implementations f1 and f2 are more than

80 times slower than the winner. f3 is the slowest version, since the use of the eval

function for such a large number of evaluations generates a huge overhead. In the case of

numexpr, the string-based expression is evaluated once and then compiled for later use;

with the Python eval function this evaluation takes place 500,000 times.