Linear Combinations In Python/numpy

greetings, I'm not sure if this is a dumb question or not. Lets say I have 3 numpy arrays, A1,A2,A3, and 3 floats, c1,c2,c3 and I'd like to evaluate B = A1*c1+ A2*c2+ A3*c3 will nu

Solution 1:

While numpy, in theory, could at any time always upgrade its internals to perform wondrous optimizations, at the present time it does not: B = A1*c1 + A2*c2 + A3*c3 will indeed produce and then discard intermediate temporary arrays ("spending" some auxiliary memory, of course -- nothing else).

B = A1 * c1 followed by B += A2 * c2; B += A3 * c3, again at this time, will therefore avoid spending some of that temporary memory.

Of course, you'll be able to tell the difference only if you're operating in an environment with scarce real memory (where some of that auxiliary memory is just virtual and leads to page faults) and for sufficiently large arrays to "spend" all real memory and then some. Under such extreme conditions, however, a little refactoring can buy you some performance.

Solution 2:

This is the idea behind numexpr (A Fast numerical array expression evaluator for Python and NumPy). You might give this package a try before compiling your own routines.