[Scipy-tickets] [SciPy] #1647: BUG: stats: _argcheck function for the erlang distribution is misnamed.
SciPy Trac
scipy-tickets@scipy....
Sat May 5 14:21:37 CDT 2012
#1647: BUG: stats: _argcheck function for the erlang distribution is misnamed.
------------------------------+---------------------------------------------
Reporter: warren.weckesser | Owner: somebody
Type: defect | Status: needs_review
Priority: normal | Milestone: Unscheduled
Component: scipy.stats | Version: 0.10.0
Keywords: |
------------------------------+---------------------------------------------
Comment(by josefpktd):
about fit. My guess (!) is that if the function is nicely behaved, then
using continuous optimization (with low precision for convergence of
parameters) should be faster than integer search without good starting
values.
Just rounding the shape parameter won't be enough, because we would have
to reestimate loc and scale (if they are not fixed) with the rounded
parameter, and most likely have to check both sides (floor and ceil).
What I tried to explain with the example of fitting the t distribution,
is, that I don't know if someone would want this. The first time I was
fitting the t-distribution with another package, I was surprised to get a
float, non-integer valued estimate for the shape (degrees of freedom). But
if we look at it just as a heavy tailed distribution, then there is no
reason to look only at integer degrees of freedom. Even in statistics we
have in some cases fractional dof.
The erlang case is a bit different because it is essentially a duplicate
of gama, so imposing integer values wouldn't prevent users from estimating
the same distribution with continuous parameter values.
--
Ticket URL: <http://projects.scipy.org/scipy/ticket/1647#comment:10>
SciPy <http://www.scipy.org>
SciPy is open-source software for mathematics, science, and engineering.
More information about the Scipy-tickets
mailing list