[Scipy-tickets] [SciPy] #497: Relative entropy calculation has sign inverted
SciPy
scipy-tickets@scipy....
Sun Sep 9 21:36:14 CDT 2007
#497: Relative entropy calculation has sign inverted
--------------------+-------------------------------------------------------
Reporter: lars | Owner: somebody
Type: defect | Status: new
Priority: low | Milestone: 0.6
Component: Other | Version: 0.5.2
Severity: minor | Keywords:
--------------------+-------------------------------------------------------
Hello,
The scipy.stats.entropy(distA, distB) method has an incorrect formula, in
that it returns the correct value, but negative instead of positive. This
is minor, but annoying.
The current implemented formula is:
S = -sum(pk * log(pk / qk), axis=0)
The actual formula should be:
S = sum(pk * log(pk / qk), axis=0)
Indeed, the returned value should never be negative, as this method is
returning.
See http://en.wikipedia.org/wiki/Kullback-Leibler_divergence for
reference.
--
Ticket URL: <http://scipy.org/scipy/scipy/ticket/497>
SciPy <http://www.scipy.org/>
SciPy is open-source software for mathematics, science, and engineering.
More information about the Scipy-tickets
mailing list