vibespatial.predicates.polygon¶

Attributes¶

`DE9IM_II`
`DE9IM_IB`
`DE9IM_IE`
`DE9IM_BI`
`DE9IM_BB`
`DE9IM_BE`
`DE9IM_EI`
`DE9IM_EB`
`DE9IM_EE`

Classes¶

`NativeConvexMaskCertificate`	Typed device-capable certificate for one convex polygonal row.
`NativeSimplePolygonalSourceCertificate`	Typed family-row certificate for simple positive-area source rings.

Functions¶

`certify_single_polygonal_convex_no_holes_gpu`(...)	Return a cached exact-sign convex/no-hole certificate.
`certify_polygonal_sources_simple_no_holes_gpu`(...)	Certify every physical source row for the convex vertex theorem.
`compute_polygonal_covered_by_single_convex_grouped_gpu`(...)	Classify all family coordinates, then reduce by source geometry.
`evaluate_de9im_grouped_device`(d_masks, predicate, *, ...)	Evaluate one device-counted DE-9IM span into shared tile output.
`evaluate_predicate_from_de9im`(→ numpy.ndarray)	Evaluate a spatial predicate from DE-9IM bitmasks.
`evaluate_predicate_from_de9im_device`(→ object)	Evaluate a spatial predicate from device-resident DE-9IM bitmasks.
`compute_rect_bounds_polygon_mask_predicates_gpu`(...)	Evaluate rectangle-bounds `intersects` and `covered_by` predicates.
`compute_polygonal_covered_by_single_mask_no_holes_gpu`(...)	Evaluate `query covered_by mask` for one polygonal mask on device.
`compute_polygonal_covered_by_pair_rows_no_holes_gpu`(...)	Evaluate polygonal `query covered_by mask` relation pairs on device.
`compute_polygonal_intersects_gpu`(→ numpy.ndarray \| None)	Evaluate polygonal `intersects` as a boolean relation-pair refine.
`compute_polygon_de9im_gpu`(→ numpy.ndarray \| None)	Compute DE-9IM bitmasks for geometry candidate pairs on GPU.
`supported_predicate_families`(...)	Return the set of family pairs supported by polygon predicate kernels.

Module Contents¶

vibespatial.predicates.polygon.DE9IM_II = 1¶

vibespatial.predicates.polygon.DE9IM_IB = 2¶

vibespatial.predicates.polygon.DE9IM_IE = 4¶

vibespatial.predicates.polygon.DE9IM_BI = 8¶

vibespatial.predicates.polygon.DE9IM_BB = 16¶

vibespatial.predicates.polygon.DE9IM_BE = 32¶

vibespatial.predicates.polygon.DE9IM_EI = 64¶

vibespatial.predicates.polygon.DE9IM_EB = 128¶

vibespatial.predicates.polygon.DE9IM_EE = 256¶

class vibespatial.predicates.polygon.NativeConvexMaskCertificate¶

Typed device-capable certificate for one convex polygonal row.

source_token: str¶

owner_id: int¶

device_state_id: int¶

row_mapping_id: int¶

family: vibespatial.geometry.buffers.GeometryFamily¶

family_row: int¶

coordinate_count: int¶

values: Any¶

convex_simple: bool¶

residency: vibespatial.runtime.residency.Residency¶

readiness: vibespatial.api._native_state.NativeStreamReadiness¶

validate(owner: vibespatial.geometry.owned.OwnedGeometryArray, state: Any, buffer: Any, *, family: vibespatial.geometry.buffers.GeometryFamily, family_row: int) → None¶

class vibespatial.predicates.polygon.NativeSimplePolygonalSourceCertificate¶

Typed family-row certificate for simple positive-area source rings.

source_token: str¶

owner_id: int¶

device_state_id: int¶

row_mapping_id: int¶

family: vibespatial.geometry.buffers.GeometryFamily¶

family_row_count: int¶

coordinate_count: int¶

values: Any¶

all_certified: bool¶

residency: vibespatial.runtime.residency.Residency¶

readiness: vibespatial.api._native_state.NativeStreamReadiness¶

validate(owner: vibespatial.geometry.owned.OwnedGeometryArray, state: Any, buffer: Any, *, family: vibespatial.geometry.buffers.GeometryFamily) → None¶

vibespatial.predicates.polygon.certify_single_polygonal_convex_no_holes_gpu(mask_owned: vibespatial.geometry.owned.OwnedGeometryArray, *, mask_family: vibespatial.geometry.buffers.GeometryFamily, family_row: int = 0) → NativeConvexMaskCertificate¶

Return a cached exact-sign convex/no-hole certificate.

The one-byte device result is an explicit planning packet. It crosses the boundary once per immutable OwnedGeometryArray generation and is then reused by every public predicate consumer of that mask.

vibespatial.predicates.polygon.certify_polygonal_sources_simple_no_holes_gpu(source_owned: vibespatial.geometry.owned.OwnedGeometryArray, *, source_family: vibespatial.geometry.buffers.GeometryFamily) → NativeSimplePolygonalSourceCertificate¶

Certify every physical source row for the convex vertex theorem.

The sufficient source domain is deliberately narrow: one simple, non-empty, positive-area ring (and one part for MultiPolygon). Anything else declines the complete-batch fast path and retains the general exact predicate. This makes invalid geometries fail closed without making a vertex-only claim about arbitrary OGC topology.

vibespatial.predicates.polygon.compute_polygonal_covered_by_single_convex_grouped_gpu(query_owned: vibespatial.geometry.owned.OwnedGeometryArray, mask_owned: vibespatial.geometry.owned.OwnedGeometryArray, *, query_family: vibespatial.geometry.buffers.GeometryFamily, mask_family: vibespatial.geometry.buffers.GeometryFamily)¶

Classify all family coordinates, then reduce by source geometry.

This is the measured direct/broadcast convex-containment shape. It emits no candidate relation and constructs no point geometry or dense group-code vector. The authoritative family coordinate offsets are consumed directly by the bounded segmented ALL reducer.

vibespatial.predicates.polygon.evaluate_de9im_grouped_device(d_masks, predicate: str, *, source_offset, logical_count, launch_capacity: int, out)¶: Evaluate one device-counted DE-9IM span into shared tile output.

vibespatial.predicates.polygon.evaluate_predicate_from_de9im(masks: numpy.ndarray, predicate: str) → numpy.ndarray¶: Evaluate a spatial predicate from DE-9IM bitmasks.

Parameters¶

masks : uint16 array of DE-9IM bitmasks predicate : one of the supported predicate names

Returns¶

bool array

vibespatial.predicates.polygon.evaluate_predicate_from_de9im_device(d_masks: object, predicate: str) → object¶: Evaluate a spatial predicate from device-resident DE-9IM bitmasks.

vibespatial.predicates.polygon.compute_rect_bounds_polygon_mask_predicates_gpu(mask_owned: vibespatial.geometry.owned.OwnedGeometryArray, rect_bounds: object, *, mask_family: vibespatial.geometry.buffers.GeometryFamily, return_device: bool = False) → tuple[object, object] | tuple[numpy.ndarray, numpy.ndarray] | None¶

Evaluate rectangle-bounds intersects and covered_by predicates.

Physical shape: a device rowset of axis-aligned rectangle bounds against one polygon mask row. The source rectangles do not need to be materialized as polygon geometry for this predicate refine; the constructive stage can still consume the original source rows for exact output assembly.

vibespatial.predicates.polygon.compute_polygonal_covered_by_single_mask_no_holes_gpu(query_owned: vibespatial.geometry.owned.OwnedGeometryArray, mask_owned: vibespatial.geometry.owned.OwnedGeometryArray, left_indices: numpy.ndarray | None = None, *, query_family: vibespatial.geometry.buffers.GeometryFamily, mask_family: vibespatial.geometry.buffers.GeometryFamily, d_left: object | None = None, d_pair_count: object | None = None, pair_capacity: int | None = None, return_device: bool = False) → numpy.ndarray | None¶

Evaluate query covered_by mask for one polygonal mask on device.

Convex no-hole masks use a cheaper one-sided proof in the kernel. Concave, multipart, and hole-bearing masks fall through to the exact polygon DE-9IM device path instead of a host-side capability branch.

vibespatial.predicates.polygon.compute_polygonal_covered_by_pair_rows_no_holes_gpu(query_owned: vibespatial.geometry.owned.OwnedGeometryArray, mask_owned: vibespatial.geometry.owned.OwnedGeometryArray, *, query_family: vibespatial.geometry.buffers.GeometryFamily, mask_family: vibespatial.geometry.buffers.GeometryFamily, d_left: object, d_right: object, d_pair_count: object | None = None, pair_capacity: int | None = None, return_device: bool = False) → numpy.ndarray | None¶

Evaluate polygonal query covered_by mask relation pairs on device.

This preserves the relation-pair physical shape for callers that already hold device row arrays. Convex no-hole masks use the cheaper one-sided proof; concave, multipart, and hole-bearing masks remain exact by falling through inside the kernel to the polygon DE-9IM predicate path.

vibespatial.predicates.polygon.compute_polygonal_intersects_gpu(query_owned: vibespatial.geometry.owned.OwnedGeometryArray, tree_owned: vibespatial.geometry.owned.OwnedGeometryArray, left_indices: numpy.ndarray | None = None, right_indices: numpy.ndarray | None = None, *, query_family: vibespatial.geometry.buffers.GeometryFamily, tree_family: vibespatial.geometry.buffers.GeometryFamily, d_left: object | None = None, d_right: object | None = None, d_pair_count: object | None = None, pair_capacity: int | None = None, d_pair_offset: object | None = None, launch_capacity: int | None = None, d_out: object | None = None, return_device: bool = False) → numpy.ndarray | None¶

Evaluate polygonal intersects as a boolean relation-pair refine.

This is narrower than a full DE-9IM matrix. It returns the boolean predicate directly and lets the kernel early-exit on boundary contact or containment, preserving device relation pairs for downstream consumers.

vibespatial.predicates.polygon.compute_polygon_de9im_gpu(query_owned: vibespatial.geometry.owned.OwnedGeometryArray, tree_owned: vibespatial.geometry.owned.OwnedGeometryArray, left_indices: numpy.ndarray | None = None, right_indices: numpy.ndarray | None = None, *, query_family: vibespatial.geometry.buffers.GeometryFamily, tree_family: vibespatial.geometry.buffers.GeometryFamily, d_left: object | None = None, d_right: object | None = None, d_pair_count: object | None = None, pair_capacity: int | None = None, d_pair_offset: object | None = None, launch_capacity: int | None = None, d_mask: object | None = None, return_device: bool = False) → numpy.ndarray | None¶

Compute DE-9IM bitmasks for geometry candidate pairs on GPU.

Supports all combinations of LINESTRING, MULTILINESTRING, POLYGON, and MULTIPOLYGON families.

When d_left / d_right are provided (device-resident CuPy int32 arrays), they are used directly instead of uploading left_indices / right_indices from host — avoiding a redundant host->device transfer when candidates are already on device.

Parameters¶

left_indices, right_indicesnp.ndarray or None: Host index arrays. May be None when d_left / d_right are provided — pair_count is derived from d_left.shape[0].
return_devicebool: When True, return the result as a device-resident CuPy uint16 array (caller takes ownership). The D->H copy is skipped entirely. Default False preserves backward compatibility (returns host np.ndarray).
d_pair_countcupy.ndarray or None: Optional device scalar naming how many leading entries in d_left / d_right are live. This preserves candidate-refine rowsets whose count is produced by a device compaction kernel without a host scalar sizing fence.
pair_capacityint or None: Launch capacity for d_left / d_right when d_pair_count is used.

Returns¶

np.ndarray | cupy.ndarray | None: uint16 DE-9IM bitmask array, or None if the family pair is not supported.

vibespatial.predicates.polygon.supported_predicate_families() → frozenset[tuple[vibespatial.geometry.buffers.GeometryFamily, vibespatial.geometry.buffers.GeometryFamily]]¶: Return the set of family pairs supported by polygon predicate kernels.