Simple Indexing

named_arrays generalizes the numpy indexing model using dicts insted of tuples. This means that indexing the array depends on the name of the axis instead of it’s position.

import named_arrays as na

Say we have a given 2D array, with axis names row and col.

a = na.arange(1, 7, "a").reshape(dict(row=2, col=3))
a

ScalarArray(
    ndarray=[[1, 2, 3],
             [4, 5, 6]],
    axes=('row', 'col'),
)

If we want to select the top-left element of a, we need to use a dict with the row and column set to zero. It would be nice if we could use keyword arguments to __getitem__ directly, but this is not supported by Python. See PEP 637 for more details.

a[dict(row=0, col=0)]

ScalarArray(ndarray=np.int64(1), axes=())

If we just want to select the first row, use a dict with only the row set to zero.

a[dict(row=0)]

ScalarArray(
    ndarray=[1, 2, 3],
    axes=('col',),
)

To select only the first two columns, we need to use a slice object. Slicing notation is not supported.

a[dict(col=slice(2))]

ScalarArray(
    ndarray=[[1, 2],
             [4, 5]],
    axes=('row', 'col'),
)

If the axis given by the item key does not exist, it is ignored. This allows vector and function indexing to work more intuitively and efficiently.

a[dict(foo=1)]

ScalarArray(
    ndarray=[[1, 2, 3],
             [4, 5, 6]],
    axes=('row', 'col'),
)

This document only describes simple indexing, but advanced indexing is also fully supported and will be discussed in later tutorials.