Slicing and indexing

Slicing and indexing are standard operations for those familiar with multidimensional array data. In Zarr, these operations are applied over "chunked" array storage, enabling users to efficiently access subsets of data, even when the data size exceeds memory capacity.

While slicing and indexing are foundational concepts as in Zarr, they are presented through a higher-level (optional) API in zarrita. This choice caters to applications that might prefer direct interaction with chunks.

You can use either @zarrita/indexing or @zarrita/ndarray to conveniently access specific data subsets without thinking about chunking details.

How to slice

To slice a zarr.Array, you must define a selection across each dimension of your data. A dimension selection can be either:

• null: all elements of the dimension.
• number: a specific (integer) index within the dimension.
• Slice: a specific subset within a dimension.

Concretely, if you have an zarr.Array of shape [512, 256, 3] and want to access 10 elements from the first dimension, all from the second, and the first from the third (resulting in a 2D array shaped [10, 256]), you'd use:

import { get, slice } from "@zarrita/indexing";
const region = await get(arr, [slice(10, 20), null, 0]);

INFO

The slice utility supports:

• slice(end)
• slice(start, end)
• slice(start, end, step)

Note that slice(null) and null are equivalent selections.

To draw a parallel, this operation in Python would look like:

region = arr[10:20, ..., 0]