Visualize task graphs¶
||Visualize several dask graphs at once.|
Before executing your computation you might consider visualizing the underlying task graph. By looking at the inter-connectedness of tasks you can learn more about potential bottlenecks where parallelism may not be possile, or areas where many tasks depend on each other, which may cause a great deal of communication.
.visualize method and
dask.visualize function work exactly like
.compute method and
except that rather than computing the result,
they produce an image of the task graph.
import dask.array as da x = da.ones((15, 15), chunks=(5, 5)) y = x + x.T # y.compute() y.visualize(filename='transpose.svg')
Note that the
visualize function is powered by the GraphViz
system library. This library has a few considerations:
- You must install both the graphviz system library (with tools like apt-get, yum, or brew)
and the graphviz Python library.
If you use Conda then you need to install
python-graphviz, which will bring along the
graphvizsystem library as a dependency.
- Graphviz takes a while on graphs larger than about 100 nodes. For large computations you might have to simplify your computation a bit for the visualize method to work well.