v1.6.5 — float dtype consistency
A small maintenance release that replaces all uses of the deprecated jnp.float_ alias with Python's built-in float across the source tree, for forward compatibility with newer JAX versions and consistency with the earlier float32 → float cleanup.
🛠 Maintenance
- Dtype consistency: Replaced
jnp.float_withfloatin:src/bbob_jax/_src/bbob.py(Gallagher precompute, Katsuura)src/bbob_jax/_src/cec2005.py(f3,_elliptic_base)src/bbob_jax/_src/registry.py(conditioned linear transform, nonsingular integer matrix sampler, linear slope makers, CEC2005 conditioned/F12 makers)src/bbob_jax/_src/utils.py(lambda_func,bernoulli_vector,cec2005_weierstrass)
- Imports: Added
Sequencetocollections.abcimport inregistry.py.
📦 Release
- Bumped version to 1.6.5 in
pyproject.tomlanduv.lock.
✅ Verification
- No functional/behavioral changes —
floatresolves to the same default JAX dtype asjnp.float_.
What's Changed
- chore: release v1.6.5 — float dtype consistency by @mpvanderschelling in #14
Full Changelog: v1.6.4...v1.6.5