You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
SymLogNorm and SymLogScale give inconsistent results.... #16376
I feel this issue is a pretty big emergency. Anyone using this norm for published work is publishing incorrect data. The fact that it’s been broken forever is pretty brutal because who knows how many papers have this error. (Hopefully few, because it’s a pretty unconventional norm).
OK, the issue is that SymLogNorm is hard coded with base np.e, whereas scale takes a variable base, but defaults to 10. I guess its arguable what the base should be by default. ping @dstansby? Did you know the norm was not base-10, but rather base-e? Should we just keep it that way for historic reasons? I don't think most people want or expect base-e...
Just a heads up that I'm going to try and tackle this again. #16391 as it stands has tests with correct and hand-calculable values, that fail for norm (without the changes in that PR) and scale. Hopefully we can get #16457 in for LogNorm, which will then allow me to derive SymLogNorm from SymLogScale and fix both at the same time.
Activity
jklymak commentedon Jan 30, 2020
Related to #16280 and #16286
jklymak commentedon Jan 30, 2020
Marking release critical because anyone who uses
SymLogNormis possibly plotting corrupted data?jklymak commentedon Jan 31, 2020
I feel this issue is a pretty big emergency. Anyone using this norm for published work is publishing incorrect data. The fact that it’s been broken forever is pretty brutal because who knows how many papers have this error. (Hopefully few, because it’s a pretty unconventional norm).
jklymak commentedon Feb 1, 2020
See #16375 (comment) for @anntzer possible path to fixing the problem.
jklymak commentedon Feb 2, 2020
OK, the issue is that SymLogNorm is hard coded with base
np.e, whereas scale takes a variable base, but defaults to 10. I guess its arguable what the base should be by default. ping @dstansby? Did you know the norm was not base-10, but rather base-e? Should we just keep it that way for historic reasons? I don't think most people want or expect base-e...jklymak commentedon Feb 2, 2020
Like the doc string even says "decades", which I don't think it correct for base-e.
dstansby commentedon Feb 2, 2020
I always read it as base 10, because it says 'decades'.
dstansby commentedon Feb 2, 2020
PR open, thanks again for tracking this down and opening the issue
dstansby commentedon Jun 26, 2020
Just a heads up that I'm going to try and tackle this again. #16391 as it stands has tests with correct and hand-calculable values, that fail for norm (without the changes in that PR) and scale. Hopefully we can get #16457 in for LogNorm, which will then allow me to derive SymLogNorm from SymLogScale and fix both at the same time.
jklymak commentedon Jan 5, 2021
This was closed by #16457 FAICT...