Saving scatterplot animations
Asked Answered
S

1

6

I've been trying to save an animated scatterplot with matplotlib, and I would prefer that it didn't require totally different code for viewing as an animated figure and for saving a copy. The figure shows all the datapoints perfectly after the save completes.

This code is a modified version of Giggi's on Animating 3d scatterplot in matplotlib, with a fix for the colors from Yann's answer on Matplotlib 3D scatter color lost after redraw (because colors will be important on my video, so I want to make sure they work).

import matplotlib.pyplot as plt
import matplotlib.animation as animation
import numpy as np
from mpl_toolkits.mplot3d import Axes3D

FLOOR = -10
CEILING = 10

class AnimatedScatter(object):
    def __init__(self, numpoints=5):
        self.numpoints = numpoints
        self.stream = self.data_stream()
        self.angle = 0

        self.fig = plt.figure()
        self.fig.canvas.mpl_connect('draw_event',self.forceUpdate)
        self.ax = self.fig.add_subplot(111,projection = '3d')
        self.ani = animation.FuncAnimation(self.fig, self.update, interval=100, 
                                       init_func=self.setup_plot, blit=True,frames=20)

    def change_angle(self):
        self.angle = (self.angle + 1)%360

    def forceUpdate(self, event):
        self.scat.changed()

    def setup_plot(self):
        X = next(self.stream)
        c = ['b', 'r', 'g', 'y', 'm']
        self.scat = self.ax.scatter(X[:,0], X[:,1], X[:,2] , c=c, s=200, animated=True)

        self.ax.set_xlim3d(FLOOR, CEILING)
        self.ax.set_ylim3d(FLOOR, CEILING)
        self.ax.set_zlim3d(FLOOR, CEILING)

        return self.scat,

    def data_stream(self):
        data = np.zeros(( self.numpoints , 3 ))
        xyz = data[:,:3]
        while True:
            xyz += 2 * (np.random.random(( self.numpoints,3)) - 0.5)
            yield data

    def update(self, i):
        data = next(self.stream)
        #data = np.transpose(data)

        self.scat._offsets3d = ( np.ma.ravel(data[:,0]) , np.ma.ravel(data[:,1]) , np.ma.ravel(data[:,2]) )

        plt.draw()
        return self.scat,

    def show(self):
        plt.show()

if __name__ == '__main__':
    a = AnimatedScatter()
    a.ani.save("movie.avi", codec='avi')
    a.show()

A perfectly valid .avi is generated by this, but it's blank for all of the four seconds except for the axes. The actual figure always shows exactly what I want to see. How can I populate the save function's plots the same way I populate a normally running animation, or is it possible in matplotlib?

EDIT: Using a scatter call in the update (without setting the bounds as in the initializer) causes the .avi to show the axes growing, showing that the data is being run each time, it's just not showing on the video itself. I am using matplotlib 1.1.1rc with Python 2.7.3.

Shoer answered 6/2, 2013 at 22:18 Comment(5)
can you fix your indentation?Animator
Whoops. I should be more careful copying and pasting. Although my source has spaces instead of tabs, so I don't quite know what happened.Shoer
and this looks ok if you watch it in a figure? What version of MPL are you using?Animator
The figure isn't a problem, the .avi is. I did note that it worked in the first paragraph, but let me edit that for emphasis. I am using version 1.1.1rc.Shoer
if you remove plt.draw() from update it breaks in the figure, which I think in a clue.Animator
A
2

Remove blit=True from FuncAnimation and animated=True from scatter and it works. I suspect that there is something going wrong with the logic that makes sure only the artists that need to be updated are updated/redrawn between frames (rather than just re-drawing everything).

Below is exactly what I ran and I got the expected output movie:

import matplotlib.pyplot as plt
import matplotlib.animation as animation
import numpy as np
from mpl_toolkits.mplot3d import Axes3D

FLOOR = -10
CEILING = 10

class AnimatedScatter(object):
    def __init__(self, numpoints=5):
        self.numpoints = numpoints
        self.stream = self.data_stream()
        self.angle = 0

        self.fig = plt.figure()
        self.fig.canvas.mpl_connect('draw_event',self.forceUpdate)
        self.ax = self.fig.add_subplot(111,projection = '3d')
        self.ani = animation.FuncAnimation(self.fig, self.update, interval=100, 
                                       init_func=self.setup_plot, frames=20)

    def change_angle(self):
        self.angle = (self.angle + 1)%360

    def forceUpdate(self, event):
        self.scat.changed()

    def setup_plot(self):
        X = next(self.stream)
        c = ['b', 'r', 'g', 'y', 'm']
        self.scat = self.ax.scatter(X[:,0], X[:,1], X[:,2] , c=c, s=200)

        self.ax.set_xlim3d(FLOOR, CEILING)
        self.ax.set_ylim3d(FLOOR, CEILING)
        self.ax.set_zlim3d(FLOOR, CEILING)

        return self.scat,

    def data_stream(self):
        data = np.zeros(( self.numpoints , 3 ))
        xyz = data[:,:3]
        while True:
            xyz += 2 * (np.random.random(( self.numpoints,3)) - 0.5)
            yield data

    def update(self, i):
        data = next(self.stream)
        self.scat._offsets3d = ( np.ma.ravel(data[:,0]) , np.ma.ravel(data[:,1]) , np.ma.ravel(data[:,2]) )
        return self.scat,

    def show(self):
        plt.show()

if __name__ == '__main__':
    a = AnimatedScatter()
    a.ani.save("movie.avi", codec='avi')
    a.show()
Animator answered 6/2, 2013 at 23:14 Comment(3)
That did the trick! And my actual project worked with that fix. Thanks!Shoer
It's interesting that removing blit helps, considering the animation.save code sets blit to false when calling _draw_next_frame. I'll add full bug reports (or at least wishlist requests) to matplotlib's site when I get the chance.Shoer
With SciPy 0.12.0, Numpy 1.7.1, and Matplotlib 1.2.1 I could not make it work. It at best would create a video with a ton of old points plotted overwritten by the new points or a blank plot. You are welcome to try the code that is linked on my video: youtube.com/watch?v=dCIOOjJ1Sk0 I finally used CamStudio and it worked better and had higher quality video output.Moonlit

© 2022 - 2024 — McMap. All rights reserved.