Line (`line`)¶

Terms¶

Line segment: A pair of vertices (A, B) connected by a straight edge. A line store holds an unordered collection of such pairs. Unlike polylines, there is no sequential ordering between segments; each segment is independent.
GEOM_LINE: The geometry type constant "line". Stored in root .zattrs under "geometry_type".
links/<delta>/: The array storing pairs of local-chunk vertex indices for each line segment in a chunk. Shape (E, 2) int32 per chunk, where E is the number of segments in the chunk. Both endpoints of a segment must lie in the same chunk (cross-chunk segments are not supported for this type).

Introduction¶

The line type stores independent line segments: pairs of vertices with no sequential relationship between segments. It is appropriate for contact sites (two endpoints of a synapse), short connectors in circuit diagrams, pairwise distance annotations, or any other dataset where the fundamental unit is a two-point segment rather than an ordered path.

line is simpler than polyline because segments do not span chunks (no cross_chunk_links/ needed) and there is no object model (no object_index/). It is therefore faster to write and read than polyline for data that genuinely consists of independent segments.

Technical reference¶

Arrays present¶

Array path	Required	Description
`vertices/`	Yes	Endpoint positions, shape `(N, D)` float32 per chunk
`vertex_fragments/`	Yes	Fragment index over `vertices/` rows
`links/<delta>/`	Yes	Segment pairs, shape `(E, 2)` int32 per chunk
`link_fragments/`	Yes (`<delta>=0`)	Fragment index over `links/0/` rows
`attributes/<name>/`	No	Per-vertex attributes

No object_index/, cross_chunk_links/, or object_attributes/ arrays.

Constraint: both endpoints in the same chunk¶

A line segment’s two vertices must fall within the same ZVF spatial chunk. Segments that cross a chunk boundary are not supported by this type; use polyline with cross_chunk_links for such data.

At write time, write_lines raises ValueError if any segment has endpoints in different chunks. To handle cross-chunk segments, pass split_cross_chunk=True, which splits the segment at the chunk boundary and inserts a midpoint vertex; this changes the geometry slightly and is not always appropriate.

Vertex ordering in `links/<delta>/`¶

Edge indices are local to the chunk (0-indexed within the chunk’s vertex slice). An edge [i, j] in chunk (cx, cy, cz) refers to vertices at positions vertices[cx, cy, cz, i] and vertices[cx, cy, cz, j].

Edges are stored in the same VG order as their source vertex (the vertex with lower array index): edges whose first endpoint is in bin 0 come first, then bin 1, etc.

Write API¶

import numpy as np
from zarr_vectors.types.lines import write_lines

# n_segments × 2 × D array: each row is a segment, columns are endpoints
segments = np.random.default_rng(0).uniform(0, 500, (10_000, 2, 3)).astype(np.float32)

# Reshape to flat vertices and edge pairs
n = len(segments)
verts = segments.reshape(n * 2, 3)            # (2n, D)
edges = np.column_stack([                      # (n, 2)
    np.arange(0, 2*n, 2),
    np.arange(1, 2*n, 2),
]).astype(np.int32)

write_lines(
    "contacts.zarrvectors",
    vertices=verts,
    edges=edges,
    chunk_shape=(200.0, 200.0, 200.0),
    bin_shape=(50.0, 50.0, 50.0),
)

Alternatively, pass a list of (A, B) tuples:

from zarr_vectors.types.lines import write_line_pairs

pairs = [(rng.uniform(0, 500, 3), rng.uniform(0, 500, 3)) for _ in range(10_000)]
write_line_pairs("contacts.zarrvectors", pairs,
                 chunk_shape=(200., 200., 200.))

Read API¶

from zarr_vectors.types.lines import read_lines

result = read_lines("contacts.zarrvectors")
print(result["segment_count"])       # int
print(result["vertices"].shape)      # (2N, D)
print(result["edges"].shape)         # (N, 2)

# Return as (N, 2, D) array of segment endpoint pairs
result = read_lines("contacts.zarrvectors", return_pairs=True)
print(result["pairs"].shape)         # (N, 2, D)

# Spatial query
result = read_lines(
    "contacts.zarrvectors",
    bbox=(np.array([0., 0., 0.]), np.array([200., 200., 200.])),
)

Relationship to `polyline`¶

line and polyline share the links/<delta>/ array schema. The distinction is:

Property	`line`	`polyline`
Segment order	Independent	Sequential (ordered path)
Cross-chunk segments	Not supported	Supported via `cross_chunk_links/`
Object index	No	Yes
Object model	No	Yes

If your data has ordered paths (e.g. vessel centrelines from which individual segment pairs were extracted), polyline preserves the ordering and enables object-level queries. Use line only for genuinely unordered, independent segment pairs.

Validation¶

L1: vertices/, vertex_fragments/, links/<delta>/, and link_fragments/ (at <delta>=0) exist.

L2: No object_index/ or cross_chunk_links/ present.

L3:

For every chunk, all edge vertex indices are in [0, N_chunk) where N_chunk is the vertex count in that chunk.
No edge has both indices equal (no self-loops).

Line (line)¶