matplotlib eventplot not shows all the binary data for big number of events #20243
Comments
|
The issue here is that the line length is the set in units of "number of sets" so a line length of 1 does not over lap with the set above or below, a line length of 2 half overlaps with its neighbors and so on. However, in this case because you have 500 sets and many fewer than 500 pixels of vertical height in each sub-plot. This means that many of the lines are less than a pixel long! For less that 1 pixel long lines I think we are falling victim to a collaboration of snapping and maybe path simplification. For a variety of reasons, as part of the rasterization process we "snap" paths to their nearest pixel. Further, as part of drawing paths we simplify them to drop any points that do not contribute to the visual path (e.g. if you have a perfectly straight line we only keep the first and last point to rasterize!). However, when you have lines that are less than a pixel long I think that if both pixels get snapped to the same pixel, we either simplify it out of existence or the Agg renderer draws nothing. The lines you are seeing in the top are when the ends get snapped to adjacent pixels and we draw a (very short) line. The visualization is further confounded by the anti-aliasing which makes the points look washed out because the line has a width that is several pixel wide, but a length that is less than 1 pixel! This results on any colored pixels only getting a fraction of the lines color composited which results in it looking like you have put a relatively high alpha on the color. By disabling snapping on import matplotlib.pyplot as plt
import numpy as np
#simulate some raster (is needed for the number of events to be high
#I noticed missing lines for n_events > 200)
data = [np.random.random(np.random.randint(10)) for i in range(500)]
fig,axes = plt.subplots(2)
axes[0].eventplot(data,linelength = 1, antialiased=True, snap=False);
axes[1].eventplot(data,linelength = 2);
To conform that they are all there, if you get this in an interactive backend you can slowly resize the window. With I am pretty convinced that if the simplification logic is turning a 2 point line into a 1 point line, that is a bug in the simplification code. If passing a path with identical points to Agg is resulting in nothing drawn that might be defendable, but also might be a bug in Agg. It would be worth trying this example with mplcairo and / or saving to a vector format to see what they do. A pragmatic fix would be to put a check in I think digging into the c++ code to check what the simplification logic is going is a good first issue for someone who is comfortable with c++ (there is no API choices to make and the code is pretty self contained), but hard (as the processing chain is implemented via templates and there is some non-trivial logic to detect if "the line has changed"!). |
|
Hey @lorenzgutierrez I would like to work on the issue |
|
Hi @RohitP2005 you are welcome to submit a pull request for any issue, but we do not assign them. Please see our contributors’ guide: |
|
understood , i will definitely look into it |
|
I am working on this issue in a group, and I have some questions. Which c++ files or functions are called upon by eventplot, and what would be the best way of debugging the code as it involves a mix of python and c++. |
|
Start digging down from |
|
To add to @timhoffm's answer, specific to Matplotlib where code is located is documented in https://matplotlib.org/devdocs/devel/contribute.html#contribute-code and I strongly suggest that you build with the compile logs https://matplotlib.org/devdocs/devel/development_setup.html#build-options |
I think this is the fundamental bug - we should not ever draw nothing if the user has asked us to draw something. I actually think that in 2025 we should do away with pixel-snapping as a default. We typically have enough resolution to see most things, even if they are fuzzy, and that is preferable to disappearing data. I'm not sure pixel snapping can be fixed for data that is less than a pixel long without forcing it to make the lines too long. |
Bug report
Bug summary
I noticed that when using eventplot to plot a big number of events, the plot doesn't shows all the binary data available, and is needed to increase the
linelenghtsparameter to increase the number of lines in the plot.Code for reproduction
Actual outcome
Expected outcome
I would expect to have all the binary data plotted in the figure
Matplotlib version
import matplotlib; print(matplotlib.__version__)): 3.3.2print(matplotlib.get_backend())):module://ipykernel.pylab.backend_inlineThe text was updated successfully, but these errors were encountered: