I implemented this another way for my own use:
https://gist.github.com/phobson/7916777

Personally, I think a vcenter kwarg would be great -- especially if it could propagate to all the methods already using vmin and vmax. That sounds like a pretty big effort, however.

@endolith @phobson If either of you want to get a PR together, we can try to get this in 1.4, otherwise, I have tagged this for 1.5

@tacaswell do you think that simply trying to add a vcenter kwarg to colors.Normalize is a good place to start?

The other question is what would this PR really being trying to accomplish? More specifically, is doing something like this notebook without having to create a totally new colormap the desired end result? http://nbviewer.ipython.org/gist/phobson/7916777

I think the simplest thing to add is @efiring suggestion to, if vcenter in passed in push out which ever of vmin and vmax is closer to make the range symmetric (to basically accomplish the upper-right graph from your notebook).

@tacaswell taking a stab at this now.

@phobson Did anything come of this?

I've got an outdated branch on a machine and I could never get all of my edge cases to pass their tests. Basically all I'm trying to do. 

Knowing what I know now about MPL's internals, I think a separate subclass would be best and could probably take a look at that tonight. The other option is to add the "vcenter" kwarg to the main Norm class. But that seems like a bad idea. Thoughts?

—
Sent from Mailbox

On Wed, Nov 26, 2014 at 2:25 PM, Thomas A Caswell
[email protected] wrote:

@phobson Did anything come of this?

Reply to this email directly or view it on GitHub:
#1806 (comment)

On one hand, a new class would be the least disruptive change, on the other hand, I think it would basically be the Normalize class with slightly different __init__.

On balance, I think a sub-class with arguments like SymNorm(center, half_width) is probably the best way to go because

• no chance of breaking existing code
• simple (minimal complicated logic, over-specified arguments)
• intuitive to find with out having to understand said complicated logic/precedence rules

#3858

Tracing related PRs, it seems like the most current related work sits in #7294

Implemented in #17709.