-
-
Notifications
You must be signed in to change notification settings - Fork 5.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
bugs stats entropy #2765
Comments
as cross-ref: a similar list with (possibly) remaining bug for higher moments and stats skew kurtosis gh-1329 |
I added the Appear OK in master Easy fix Needs further investigation For some of the distributions, there is a known explicit formula for the entropy that is not used (e.g.
|
Thanks for updating this. ksone, and kstwobign can be ignored. |
After merging gh-2774, this is what's left:
|
It can probably be closed now: Warren explained truncnorm and reciprocal fails are explained above, ksone/kstwobign are special anyway, and ncf seems to have an accuracy issue, special-cased in https://github.com/scipy/scipy/blob/master/scipy/stats/tests/test_continuous_basic.py#L428 |
@EvgeniBurovski Can you open a new issue for numerical problems in ncf? I don't have an overview for which methods the numerical problems are serious. Then we can close this. As we get closer to being bugfree, I would prefer issues for individual distributions that still need attention (instead of these generic entropy or moments don't work issues.) |
@josef-pkt done. (not sure is any of these is high priority) |
Thanks, closing. |
running the script at the bottom with python 3.3 and scipy 0.12.0
shows the following differences between the reported entropy and the one calculated by numerical integration (generic code)
This should have fewer false positives than the tests I did before with comparison to a large random sample (theoretical entropy versus sample entropy).
distributions raising an exception
distribution with more than 0.1 relative tolerance
last value is what the distributions return
second to last value is from numerical integration
(not shown: a very large number of warnings)
Notes:
The text was updated successfully, but these errors were encountered: