From: Michael Meyer (spyqqqdia_at_[hidden])
Date: 2003-02-27 00:27:40
Statistics functionality would be extremely
The idea would be to duplicate the basic notions of
the mathematical theory as C++ classes.
The conceptual hyrarchy (bottom to top) is as follows:
1. Probability distribution (implementation: random
2. Random Object (Random variable, random vector)
implementation: this is your "stats class"
3. Stochastic process.
This thread places to much importance on the "stats
class" as a container of data. Obviously this view
results from the fact that most users work from
preexisting data sets which are read into a container.
This is what is called an empirical distribution and
could be handled as special case of a random variable
-- an empirical random variable.
The case of the preexisting data set is a fairly
trivial case. More generally the data are generated by
some procedure (special case: reading from the
preexisting data set).
The procedure could be more interesting than this.
For example in Monte Carlo simulation one often
computes the path of some stochastic process and then
generates a data point as a functional (deterministic
function) of this path.
The notion of a random variable should capture all
this. I have already implemented some useful
abstractions in Java. See
I am working on C++ now and will gladly contribute
code to Boost.
Do you Yahoo!?
Yahoo! Tax Center - forms, calculators, tips, more
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk