v2.0.7

Fixed setuptools packaging issue

Release v2.0.6

Packaging

• Capped setuptools to address pkg_resources deprecation issues.

Release v2.0.5

Changes

• Benchmarking: Updated the Attentive kernel benchmark to use tf.Adam for better performance, and added mutual information (MI) to benchmark outputs.
• Attentive kernel: Skip redundant representation computation when inputs match for faster execution.
• Metrics: Added mean squared error (MSE).
• Datasets: Added support for .npy datasets.
• TSP: Removed the planner-level time limit argument and improved solver behavior/options (initial path support, solution limits, and running the secondary solver only when a time limit is provided).

Packaging

• Capped setuptools to address pkg_resources deprecation issues, tightened dependency constraints (including TensorFlow-related requirements), and added numba.

Release v2.0.4

Key Updates

• Fix default float type casting

Release v2.0.3

Key Updates

• Improve random seed pass-throughs

Release v2.0.2

Key Updates

• Add sanity check to catch inf values in the objective function output for CMA
• Update kernel function naming convention to match GPFlow
• Update non-stationary kernel function tutorial to use get_kernel and get method interface
• Improve error descriptions for get_kernel, get method, and get objective

Release v2.0.1

Key Updates

• Added project_waypoints method to project waypoints to be within the environment
• Added get_kernel method to get a kernel with a string
• Added new optimization objectives (compatible with CMA, BayesianOpt, and GreedyObjective methods)
  • A-Optimal Design
  • B-Optimal Design
  • D-Optimal Design
• Updated mutual information methods to support caching
• Added Schur complement-based mutual information
• Added new method DifferentiableObjective that can optimize any objective using gradient-based approaches

Release v2.0.0

Key Updates

• Added new unified interface for all SP/IPP methods
  • Allows getting and running any method with sgptools.methods.get_method (e.g., ContinuousSGP, CMA)
  • Allows selecting the optimizer with a string (e.g., scipy.L-BFGS-B, tf.Adam)
  • Allows changing the backend objective function with a string (e.g., MI, SLogMI)
• Added new Dataset class for cleaner data management
• Improved code readability with type hints
• Improved documentation

Release v1.2.0

Key Updates

• Added new Attentive Non-stationary Kernel
• Fixed Neural Kernel Non-stationary Kernel
• Updated SGPR's update method to support all kernel functions

Release v1.1.8

Key Updates

• Add option to skip inducing variables update in SSGP