Apr 28
So now that we're at the end of the term, I thought we could
discuss general approaches the underlying theme of the course:
how do you write code to solve a given problem that will run fast?
making it faster
0a. choose algorithm
0b. get it to work at all, and understand run time
1. profiling loop:
test it
profile it, find out which parts are slow
rewrite slowest parts
2. memoization
I give a simple example;
3. C for slow bits:
details vary with language you're embedding in;
from python one typically creates a "setup.py"
file and let python itself compile the C code.
Usually you end up writing a "wrapper" around the C.
Or you can use PyInline (!) (Though it's old.)
4. can this problem run concurrently ?
a) Will the problem let you do that?
b) If so, what do you have access to? multiple cpus? machines?
See e.g. http://wiki.python.org/moin/ParallelProcessing
5. more cpu/memory help ?
See e.g. www.picloud.com
Also:
Has someone else done the problem (or part of it) for you?
http://nodebox.net/code/index.php/Graph
Locality : in memory (don't thrash!), in access time (caches!)
... particularly in C code.
physical units
N-body example
