Sometimes problems don’t fit into one of the collections like dask.array or dask.dataframe. In these cases, users can parallelize custom algorithms using the simpler dask.delayed interface. This allows one to create graphs directly with a light annotation of normal python code.

>>> x = dask.delayed(inc)(1)
>>> y = dask.delayed(inc)(2)
>>> z = dask.delayed(add)(x, y)
>>> z.compute()
>>> z.vizualize()
simple task graph created with dask.delayed