nbagg backend incredibly laggy in IPython notebook

[alexanderwhatley]

• Matplotlib version, Python version and Platform (Windows, OSX, Linux ...)
  Windows 10, Matplotlib 1.5.1, Python 3.5.2

I wrote some code for allowing the user to draw rectangles in a blank matplotlib graph using the nbagg backend in an IPython notebook, and it is incredibly laggy. Is this a current limitation, or is this perhaps a bug with the library? Here is the code I have:

import matplotlib
matplotlib.use("nbagg")
import matplotlib.pyplot as plt
from matplotlib.patches import Rectangle
from ipywidgets.widgets import Button
from IPython.display import display

class Test(object):
    def __init__(self):
        plt.ion()
        self.figure = plt.figure()
        self.ax = self.figure.gca()
        self.rect = Rectangle((0,0), 0, 0, color = '0', fill = False)
        self.x0 = None
        self.y0 = None
        self.x1 = None
        self.y1 = None
        # Graph interaction
        self.ax.add_patch(self.rect)
        self.ax.figure.canvas.mpl_connect('button_press_event', self.on_press)
        self.ax.figure.canvas.mpl_connect('button_release_event', self.on_release)
        self.ax.figure.canvas.mpl_connect('motion_notify_event', self.on_motion)

    def on_motion(self, event):
        if self.is_pressed:
            self.x1 = event.xdata
            self.y1 = event.ydata

            self.rect.set_width(self.x1 - self.x0)
            self.rect.set_height(self.y1 - self.y0)
            self.rect.set_xy((self.x0, self.y0))
            self.rect.figure.canvas.draw()

    def on_press(self, event):
        self.is_pressed = 1
        self.x0 = event.xdata
        self.y0 = event.ydata
        self.rect.set_visible(True)

    def on_release(self, event):
        self.is_pressed = 0
        self.ax.scatter(self.x_inside, self.y_inside, c = "red")
        self.figure.canvas.draw()
        self.rect.set_visible(False)
        self.rect.figure.canvas.draw()


test = Test()

[tacaswell]

Remove all of the explicit calls to draw (or at least change them to draw_idle).

Please try the 2.0.0b3 beta, there are some speed improvements on that branch as well.

The killer is probably fully re-drawing the figure on every mouse motion. The toolbar has a draw_rubberband (which is what zoom uses) which draws the rectangle using the 'native' tools and is much more performant.

[watkinrt]

Does anyone know what makes the nbagg back so much slower than the other aggs with respect to event handling? I'm trying to do a similar draggable (motion_notify_event) interaction, but the latency makes it almost unusable. I've tried using blitting and draw_idle commands to speed things up, but they only help a little (at least, using Matplotlib 2.0.0).

[tacaswell]

I suspect the latency is because the communication happening over a web-socket rather than within a process.

[thomasaarholt]

Is there anyone working on improving speeding up the notebook backend? The notebook backend is wonderful, but on Windows it is quite slow. I find that it is slower here than it is on Linux.

[tacaswell]

Having someone pick up and finish #4290 may help with this.

[github-actions]

This issue has been marked "inactive" because it has been 365 days since the last comment. If this issue is still present in recent Matplotlib releases, or the feature request is still wanted, please leave a comment and this label will be removed. If there are no updates in another 30 days, this issue will be automatically closed, but you are free to re-open or create a new issue if needed. We value issue reports, and this procedure is meant to help us resurface and prioritize issues that have not been addressed yet, not make them disappear. Thanks for your help!