
On Mon, Jun 21, 2010 at 8:30 PM, Jonathan Castello <twisolar@gmail.com> wrote:For reducing numerical error when summing numbers, see algorithms like
> It's really just the average of two numbers, (x+y)/2. An averaging
> function isn't such a bad idea, particularly if it takes an arbitrary
> number of parameters (like Jeff Pohlmeyer's implementation in this
> thread).
in Python's math.fsum [1] and [2].
This version of mean is less efficient but supports incremental
computation and is less prone to overflow (not that overflow would
likely happen for floating point numbers):
local function mean(...)
local mean = 0
for i=1,select('#', ...) do
mean = mean + (select(i, ...)  mean) / i
end
return mean
end
[1] http://docs.python.org/library/math.html#math.fsum
[2] http://en.wikipedia.org/wiki/Compensated_summation