recipe

A recipe for an array which is read a member at a time.

A Recipe holds no data. It holds members, and each member says “this window of this source goes in this box of the output”. Everything is positional: axis numbers and sample indices, never dimension names, coordinates or units.

Many arrays share one RecipeTable, which owns the storage; an array is a view of one of its rows. Members of an array are stored together, in canonical placement order, so slicing, joining and rechunking are vectorized over members and never open a file.

Examples

import numpy as np
from dascore.core.source import ArraySource
from dascore.lazy.recipe import Recipe, concat

left = Recipe.from_source(ArraySource.full((4, 3), 1.0))
right = Recipe.from_source(ArraySource.full((4, 3), 2.0))
array = concat([left, right], axis=0)
assert array.shape == (8, 3) and len(array) == 2
assert np.array_equal(array[0:5].load()[:4], np.ones((4, 3)))

Functions

Name	Description
concat	Join arrays end to end along one axis, in one pass over their members.
stack	Join arrays along a new axis, in one pass over their members.

Classes

Name	Description
Recipe	An array which says where each of its members is read from.
RecipeTable	The storage many recipes share.