Module sparse 

Sparse arrays: the storage kind behind J’s $..

A sparse array has the shape of the array it stands for and holds only the positions that differ from one repeated element — the SPARSE ELEMENT, which is zero for anything sparsify makes. Some of the axes are stored sparsely and the rest are dense, so one stored entry is a whole cell over the dense axes: an index row naming the entry’s position along the sparse axes, and that cell’s elements. When every axis is sparse — what $. y always produces — the cell is a single element and the array is the familiar list of coordinates and values.

The shape and the stored values live on the Array itself: shape is the LOGICAL shape and data is the stored cells end to end, so an array of sparse doubles reports the same dtype and formats its values through the same code a dense one does. Everything else is in Sparse, which the array carries behind an Arc.

Only $. itself, the display and ": read the stored form. Every other verb receives Array::densified, which is semantically exact and says nothing about how fast it is; the caveat is written down in docs/status.md.

Structs

Sparse

What a sparse array holds besides its shape and its stored cells.

Functions

attributes

The shape, the sparse axes and the sparse element, each boxed. This is _1 $. y.

compress

8 $. y: the same array with every stored entry whose cell is entirely the sparse element dropped. Amending a stored position back to the fill leaves the entry behind; this is what removes it.

create

1 $. y: a new sparse array with nothing stored in it. y is the shape, or a boxed shape ; axes, or a boxed shape ; axes ; element. Left to itself the whole shape is sparse and the element is a float zero, which is what J’s own bare form gives.

fill_of

The sparse element on its own, as an atom. This is 3 $. y.

indices_of

The stored index rows as an integer table: one row per entry, one column per sparse axis. This is 4 $. y.

sparsify

$. y: the dense array y in sparse form, every axis sparse and zero the sparse element. A sparse argument comes back unchanged, and a scalar stays dense — there is no axis to store it along.

values_of

The stored cells as an ordinary array: one leading axis of entries, then the cell’s own shape. This is 5 $. y.