## numexpr - Fast Evaluation in Python

From: andrew cooke <andrew@...>

Date: Sun, 15 Apr 2012 20:01:33 -0300

I just stumbled across this while wondering about writing something similar
myself (this is much more professional and complete than I was planning) - it
compiles expressions with a Python context into operations for a C-based SIMD
VM.  The result is a "compiled" operation that can give large speedups when
applied (mapped) to arrays of data.

This is, in a sense, the "dual" of numpy, in that numpy provides operations
which treat each node in the ast of an expression as a vector, while numexpr
treats the entire tree as a single operation (which is then applied to
vectors).

I'm not explaining this very well....  you would be better reading

Andrew

### Also, Numba

From: andrew cooke <andrew@...>

Date: Mon, 23 Apr 2012 00:28:39 -0300

Linking this so that if I am searching for fast numerics on Python I can find
it easily - it's an numy-aware compiler to LLVM.

https://github.com/ContinuumIO/numba

There's also Julia, which is a completely different language, but which
targets similar problems:

http://julialang.org/

Andrew