optiml 1.8

Release 1.8

• ship the ML-CUP19 dataset inside the package (package_data) so it travels with pip
• make the example notebooks runnable on Google Colab (pip-install setup cell;
  ML_report resolves its data from the package or the GitHub raw URL) + Open in Colab badge
• add pandas to the requirements so the notebooks run on Binder
• documentation fixes: render the constructor parameters and the algorithm
  descriptions (KKT systems, update rules) on the docs site, wrap everything to 120 cols

optiml 1.7

Release 1.7

• L-BFGS (limited-memory BFGS) line-search optimizer
• generic lower bound: box-constrained optimizers now solve lb <= x <= ub (no perf cost), not just 0 <= x <= ub
• SMO: fixed the linear-kernel classifier weight-vector bug; verified against the cvxopt QP reference
• certified primal optimum f_star computed directly via cvxpy
• uniform reST docstrings with LaTeX-rendered math across the library
• GitHub Pages documentation site (https://dmeoli.github.io/optiml)
• GitHub Actions CI on Python 3.9-3.12 (replacing Travis)
• modern numpy/scipy/scikit-learn/cvxpy compatibility fixes
• rigorous, portable test suite + higher coverage