Release Notes#

tinygp 0.3.0 (2024-01-05)#

Features#

  • Added a more robust and better tested implementation of the CARMA kernel for use with the QuasisepSolver. (#90)

  • Switched all base classes to equinox.Module objects to simplify dataclass handling. (#200)

Bugfixes#

  • Fixed use of jnp.roots and np.roll to make CARMA kernel jit-compliant. (#188)

tinygp 0.2.4 (2023-09-29)#

Features#

  • Removed __post_init__ checks after kernel construction to avoid extraneous errors when returning kernels out of jax.vmap’d functions. (#148)

  • Added Zenodo data to improve citation tracking. (#151)

Bugfixes#

  • Fixed syntax for vmap of flax modules in transforms tutorial. (#159)

Improved Documentation#

  • Fixed incorrect definition of “spectral mixture kernel” in the custom kernels tutorial. (#143)

  • Unpinned the docs theme version to fix release compatibility with recent versions of setuptools. (#153)

  • Added past contributor metadata to .zenodo.json. (#154)

  • Clarified in documentation that sigma argument is optional in quasisep kernels. (#176)

Misc#

tinygp 0.2.3 (2022-10-31)#

Features#

  • Removed deprecation warning from predict method and wrapped it in a jax.jit in order to support interactive use. (#120)

  • Added check for sorted input coordinates when using the QuasisepSolver; a ValueError is thrown if they are not. (#123)

Bugfixes#

  • Fixed incorrect definition of observation_model for Celerite kernel. (#88)

  • Fixed FutureWarning by updating tree_map to tree_util.tree_map. (#114)

  • Fixed issue when tree structure and shape of X_test input to condition was incompatible with the initial input. (#119)

  • Fixed bug where the gradient of the L2 distance would return NaN when the distance was zero. (#121)

  • Fixed behavior of DotProduct kernel on scalar inputs. (#124)

tinygp 0.2.2 (2022-04-20)#

Bugfixes#

  • Fixed dangling numpy operation in quasiseparable tree map. (#81)

tinygp 0.2.1 (2022-03-28)#

Features#

  • Renamed elements of quasiseparable kernels, and added support for modeling derivative observations with these kernels. (#58)

  • Added more flexible noise models: diagonal, banded, or dense. (#59)

  • Added tinygp.kernels.quasisep.CARMA kernel to implement CARMA models. (#60)

  • Added a minimal solver based on Kalman filtering to use as a baseline for checking the performance of the tinygp.solvers.QuasisepSolver. (#67)

Bugfixes#

  • Fixed exception when conditioning with quasiseparable solver, since quasisep kernels are not hashable. (#57)

Improved Documentation#

  • Added a new tutorial describing how to model multiband and derivative observations using quasiseparable kernels. (#58)

  • Add more details to Deep Kernel learning tutorial, showing comparison with Matern-3/2 kernel and the transformed features. (#70)

tinygp 0.2.0 (2022-03-03)#

Features#

Improved Documentation#

Deprecations and Removals#

  • Breaking change: Removed existing condition method and deprected predict method. (#32)