matplotlib-benchmark

Problem

Problem is that I need to draw many pictures in a for loop, with different data each time. Using matplotlib.pyplot interface is far too slow. or I didn't do it the right way. Besides, the matplotlib tutorials and docs are not very specific about the object-oriented API. So I did a lot of Googles and Stackoverflows, then I came up with this little benchmark of different plotting interface test. Speed is the very first consideration.

Code is in bench.py.

Result

It seems that the right way to do what I need is: Do not redraw everything. Just update data each time, as the function draw3 does in bench.py.

Surprisingly the first two methods don't differ too much when using Agg backend. While using Qt5Agg backend, draw1 will beat draw2 by about 15%. Almost for all three methods except draw1, Agg beats Qt5Agg.

Function	Agg time	Qt5Agg time
draw1	8.235049673209932	7.974657052376608
draw2	8.106147522373316	9.609950379663687
draw3	5.5861171293730365	5.752674005524984

Note: time unit in the table is seconds.

Ref

1. When to use cla(), clf() or close() for clearing a plot in matplotlib? - Stack Overflow
2. python - Create a figure that is reference counted - Stack Overflow
3. python - How to update a plot in matplotlib? - Stack Overflow