Array — Dask documentation (original) (raw)

Array¶

Dask Array implements a subset of the NumPy ndarray interface using blocked algorithms, cutting up the large array into many small arrays. This lets us compute on arrays larger than memory using all of our cores. We coordinate these blocked algorithms using Dask graphs.

Design¶

Dask arrays coordinate many NumPy arrays (or “duck arrays” that are sufficiently NumPy-like in API such as CuPy or Sparse arrays) arranged into a grid. These arrays may live on disk or on other machines.

New duck array chunk types (types below Dask onNEP-13’s type-casting hierarchy) can be registered viaregister_chunk_type(). Any other duck array types that are not registered will be deferred to in binary operations and NumPy ufuncs/functions (that is, Dask will return NotImplemented). Note, however, that any ndarray-like type can be inserted into a Dask Array usingfrom_array().

Common Uses¶

Dask Array is used in fields like atmospheric and oceanographic science, large scale imaging, genomics, numerical algorithms for optimization or statistics, and more.

Scope¶

Dask arrays support most of the NumPy interface like the following:

• Arithmetic and scalar mathematics: +, *, exp, log, ...
• Reductions along axes: sum(), mean(), std(), sum(axis=0), ...
• Tensor contractions / dot products / matrix multiply: tensordot
• Axis reordering / transpose: transpose
• Slicing: x[:100, 500💯-2]
• Fancy indexing along single axes with lists or NumPy arrays: x[:, [10, 1, 5]]
• Array protocols like __array__ and __array_ufunc__
• Some linear algebra: svd, qr, solve, solve_triangular, lstsq
• …

However, Dask Array does not implement the entire NumPy interface. Users expecting this will be disappointed. Notably, Dask Array lacks the following features:

• Much of np.linalg has not been implemented. This has been done by a number of excellent BLAS/LAPACK implementations, and is the focus of numerous ongoing academic research projects
• Arrays with unknown shapes do not support all operations
• Operations like sort which are notoriously difficult to do in parallel, and are of somewhat diminished value on very large data (you rarely actually need a full sort). Often we include parallel-friendly alternatives like topk
• Dask Array doesn’t implement operations like tolist that would be very inefficient for larger datasets. Likewise, it is very inefficient to iterate over a Dask array with for loops
• Dask development is driven by immediate need, hence many lesser used functions have not been implemented. Community contributions are encouraged

See the dask.array API for a more extensive list of functionality.

Execution¶

By default, Dask Array uses the threaded scheduler in order to avoid data transfer costs, and because NumPy releases the GIL well. It is also quite effective on a cluster using the dask.distributed scheduler.