New style for fast plotting, updated performance docs #8957
Conversation
|
I tried looking for the "optimal" chunksize and it turns out that (at least on my laptop) the time per edge is minimized with a chunksize of ~1000 (see https://gist.github.com/anntzer/9192485c025933cc233f3a5dc74de936) -- which already seems quite large (even though it is possible to construct contrieved examples, in general I believe that a chunksize larger than the number of pixels (or more specifically, number-of-pixels / simplify-threshold) should cause only few visible artefacts (beyond what path simplification already caused)). Interestingly, mpl_cairo also exhibits a minimum around 1000 edges, and the time per vertex grows much more sharply after that. |
| pixel perfect plots, then a value of ``1.0`` can be safely used. If you | ||
| want to make sure your plot reflects your data *exactly*, then you should | ||
| set ``path.simplify`` to false and/or ``path.simplify_threshold`` to ``0``. | ||
| Currently, matplotlib defaults to a convervative value of ``1/9``. |
There was a problem hiding this comment.
I think a version of this information should be in the core path documentation so that it is easier to stumble upon.
There was a problem hiding this comment.
@has2k1 added in most recent commits
b2d135b to
ddcbbeb
Compare
tutorials/introductory/usage.py
Outdated
| # Using the *fast* style | ||
| # ---------------------- | ||
| # | ||
| # The *fast* style that can be used to automatically set |
tutorials/introductory/usage.py
Outdated
| # # Setup, and create the data to plot | ||
| # y = np.random.rand(100000) | ||
| # y[50000:] *= 2 | ||
| # y[np.logspace(1,np.log10(50000),400).astype(int)] = -1 |
ddcbbeb to
d3f6be4
Compare
d3f6be4 to
3c27100
Compare
|
Bump. Any desire to put this in? Or is it just adding noise to the package? |
|
Thanks for bumping this! |
PR Summary
PR Checklist