Am 2013-03-08 um 22:54 schrieb Mark M <slashing75@...>:
> This seems to be a fairly common question for those who are just starting with
> SimPy. It's actually a part of a bigger question:
> "How do I run experiments in SimPy?"
> Which can then be broken down into several more questions:
> "How do I run multiple replications of my model to achieve statistical significance?"
> "How do I run multiple treatments?"
> "How do I run multiple experiments?"
This is something we should/will cover with SimPy 3’s documentation.
> It's not something that we have discussed much in the community, and I suspect
> that most of our implementations would look very similar. It is more of a python
> and experiment design issue than a SimPy issue, but running multiple replications,
> treatments and experiments is such a common thing that we might want to at least
> provide a default way to do it.
Experiment design is a whole topic of its own an nothing we can and should
cover in depth, see http://en.wikipedia.org/wiki/Taguchi_methods for example.
> - What and how we would like to output from replications and treatments.
> This mixes with the issue of "a standard set of performance metrics" which I
> think was merged with Issue 17 (Monitors). The ability for the user to record
> values of interest within a replication would have to be implemented first. The
> next feature request is usually to facilitate calculating statistics on data across
> replications (another common task) but whether or not that falls into the scope
> of SimPy is up for debate. That might be best off left to the user and statistics
> packages like R, but could certainly be made easier if we created a set of
> accomodating output backends.
Please take a look at https://bitbucket.org/ssc/simpy/commits/all for the
current monitoring draft. Next thing I’ll add is an example on how to analyse
the data collected from a resource.
However, SimPy itself won’t provide any analysis methods, because they usually
are too use case specific and are depending on the monitoring backend you
choose (e.g., python lists vs. numpy arrays vs. sql tables vs. excel).
SimPy will help you to collect and store your data, but what you do with it is
completely up to you. :-)