Conversion Policy

Context-bound objects expose convert(new_ctx). Conversion is explicit and target-context driven: the requested context controls backend, dtype, and check_level.

What moves

Spaces are reconstructed in the target context. Shape, structure, tolerances, and geometry definitions are preserved. Geometry data that stores arrays, such as WeightedInnerProduct.weights, is converted with target.asarray.

Operators are reconstructed in the target context when they implement _convert. Matrix-backed operators convert stored dense or sparse matrices. Algebraic operators convert their operands. Matrix-free operators preserve their Python callables and rebuild only their domain, codomain, and context; the callables must already be valid for the target backend.

What may fail

Conversion may fail when the target backend lacks a required sparse format, when dtype conversion is unsupported, when backend-specific callables cannot run on the target arrays, or when structural validation rejects the converted data. Optional backends are available only when their packages are installed.

Backend differences

NumPy conversion creates NumPy arrays and SciPy sparse objects. JAX conversion creates JAX arrays and follows JAX dtype configuration, including jax_enable_x64. Torch conversion creates tensors and follows PyTorch dtype and device semantics; SpaceCore does not silently move tensors across devices. CuPy conversion creates CuPy arrays and CuPy sparse objects when CuPy is installed.

Example

import numpy as np
import spacecore as sc

src = sc.Context(sc.NumpyOps(), dtype=np.float32)
dst = sc.Context(sc.NumpyOps(), dtype=np.float64)
X = sc.DenseCoordinateSpace((2,), src)
Y = X.convert(dst)
print(Y.dtype == np.dtype("float64"))

Expected output:

True