MNT: Introduce cmap(..., by_index=...) parameter #30548
Conversation
This allows to explicitly state whether inputs should be handled as normalized values [0, 1] or indices into the colormap. The value `by_index='auto'` is the current default and makes the interpretation dependent on the data type. Resulting in surprising behavior matplotlib#28198. It is planned to deprecate this in a follow-up PR and switch to `by_index=False` eventually. The standard case of float inputs will not be affected by this change. But users that want to pass indices, will then have to explicitly use `by_index=True`.
e59d0cb to
7e84272
Compare
|
I don't really think #28198 is actually a problem and would leave the behavior as is, but if we're really going to change this (I'm only -0 but nowhere close to blocking) then I'd say that indexing should just be a new operation ( |
|
I think #28198 is a footgun that should be removed. Needing the upper-limit color of a colormap is not completely unlikely, and intuition suggests Ideally, Since for the time being we need indexing capability, at least I want to make that explict and not decide call behavior based on the data type. You have two options for this, an additional paramter The |
This allows to explicitly state whether inputs should be handled as normalized values [0, 1] or indices into the colormap.
The value
by_index='auto'is the current default and makes the interpretation dependent on the data type. Resulting in surprising behavior #28198. It is planned to deprecate this in a follow-up PR and switch toby_index=Falseeventually. The standard case of float inputs will not be affected by this change. But users that want to pass indices, will then have to explicitly useby_index=True.Note: We currently rely on the auto detection for
BoundaryNorm. That will be handled by special casingBoundaryNormin the colorizer pipeline, essentially switchingcmap(norm(data))tocmap(norm(data), by_index=isinstance(norm, BoundaryNorm)). It's better to do the special casing there rather than the type-based autodetection inColormap.__call__. Hopefully, we can additionally get rid of that special case through #21911 (comment), but that's a separate discussion and its result does not impact the meaningfulness of this PR.