[SciPy-user] TimeSeries - Questions Re lib/moving_funcs.py
Tue Dec 4 22:25:17 CST 2007
I have comments/questions regarding this module:
* Why do the comments in the source code indicate that many of the functions are intended only for 1-D arrays (mov_var for example)?
A very brief look at the source doesn't indicate why this should be. Also, throwing some quick example code together suing 2-D arrays generates results. The code does not seem to bomb.
* The covariance related code is constructed in a way I would not expect.
Specifically, most of the functionality appear as facade functions fed into _mov_var_stddev. This seems to be a multi-purpose function in the module. Wouldn't it be more natural to define a general covariance function (ignoring bias or span) with pseudo-code:
cov(x,y) = [x - ave(x)]*[y - ave(y)]/n
The rest of the functions fall out of this like:
var(x) = cov(x,x)
stddev(x) = sqrt[var(x)]
corr(x,y) = cov(x,y)/stddev(x)/stddev(y)
Then add-ons like regression coefficients and zscores become trivial.
* This is a really nice package! Look forward to seeing it grow.
Looking for last minute shopping deals?
Find them fast with Yahoo! Search. http://tools.search.yahoo.com/newsearch/category.php?category=shopping
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SciPy-user