Bag

Bag¶

Dask Bag implements operations like map, filter, fold, and groupby on collections of generic Python objects. It does this in parallel with a small memory footprint using Python iterators. It is similar to a parallel version of PyToolz or a Pythonic version of the PySpark RDD.

Examples¶

Visit https://examples.dask.org/bag.html to see and run examples using Dask Bag.

Design¶

Dask bags coordinate many Python lists or Iterators, each of which forms a partition of a larger collection.

Common Uses¶

Dask bags are often used to parallelize simple computations on unstructured or semi-structured data like text data, log files, JSON records, or user defined Python objects.

Execution¶

Execution on bags provide two benefits:

Parallel: data is split up, allowing multiple cores or machines to execute in parallel
Iterating: data processes lazily, allowing smooth execution of larger-than-memory data, even on a single machine within a single partition

Default scheduler¶

By default, dask.bag uses dask.multiprocessing for computation. As a benefit, Dask bypasses the GIL and uses multiple cores on pure Python objects. As a drawback, Dask Bag doesn’t perform well on computations that include a great deal of inter-worker communication. For common operations this is rarely an issue as most Dask Bag workflows are embarrassingly parallel or result in reductions with little data moving between workers.

Shuffle¶

Some operations, like groupby, require substantial inter-worker communication. On a single machine, Dask uses partd to perform efficient, parallel, spill-to-disk shuffles. When working in a cluster, Dask uses a task based shuffle.

These shuffle operations are expensive and better handled by projects like dask.dataframe. It is best to use dask.bag to clean and process data, then transform it into an array or DataFrame before embarking on the more complex operations that require shuffle steps.

Known Limitations¶

Bags provide very general computation (any Python function). This generality comes at cost. Bags have the following known limitations:

By default, they rely on the multiprocessing scheduler, which has its own set of known limitations (see Shared Memory)
Bags are immutable and so you can not change individual elements
Bag operations tend to be slower than array/DataFrame computations in the same way that standard Python containers tend to be slower than NumPy arrays and Pandas DataFrames
Bag’s groupby is slow. You should try to use Bag’s foldby if possible. Using foldby requires more thought though

Name¶

Bag is the mathematical name for an unordered collection allowing repeats. It is a friendly synonym to multiset. A bag, or a multiset, is a generalization of the concept of a set that, unlike a set, allows multiple instances of the multiset’s elements:

list: ordered collection with repeats, [1, 2, 3, 2]
set: unordered collection without repeats, {1, 2, 3}
bag: unordered collection with repeats, {1, 2, 2, 3}

So, a bag is like a list, but it doesn’t guarantee an ordering among elements. There can be repeated elements but you can’t ask for the ith element.

Compatibility with numpy functions

Create Dask Bags

Bag

Contents