NEWS

---

rclsp 2.0.0

Changes

• Achieved alignment of the R implementation with the Python implementation.

Bug fixes

• Fixed a variable-shadowing bug in .solve.instance() where the model block M was overwritten by the internal Bott-Duffin Gram matrix during the pseudoinverse step.

rclsp 1.1.0 (2026-06-07)

Changes

• Updated the minimum R version to R 4.3 to match CVXR 1.8.x requirements.
• Restricted the supported CVXR version range to 1.8.x.

Bug fixes

• Fixed AP/TM refinement canonicalization by preserving the original problem type during iterative refinement.
• Reused the canonical right-hand-side vector during refinement to keep dimensions consistent across repeated calls to canonize.clsp().

rclsp 1.0.0 (2026-05-16)

Changes

• Preserved rcond in stored results.
• Added custom .cond() function with a relative singular-value cutoff.
• Added cond_tolerance argument to clsp().
• Used Euclidean/Frobenius norms for corr() changes in zhat, z, and x.
• Fixed a bootstrap bug in ttest().

Bug fixes

• Improved alpha handling in clsp() to correctly support numeric scalars, numeric vectors, and NULL.
• Fixed candidate alpha selection when non-finite values are supplied.
• Restricted automatic alpha selection to cases where the final convex correction step is enabled.

rclsp 0.5.0 (2026-03-09)

Bug fixes

• Updated CVXR integration for CVXR 1.8.x compatibility.

rclsp 0.4.0 (2026-02-19)

Bug fixes

• Fixed potential projector Z dimension mismatch under repeated solve() calls.

rclsp 0.3.0 (2026-01-30)

Bug fixes

• Achieved full numerical parity between R and Python implementations.
• Replaced MASS::ginv() with explicit SVD-based pseudoinverse logic.
• Aligned Bott–Duffin tolerance handling with NumPy semantics.
• Corrected row-wise reconstruction of X from the solution vector z.
• Fixed zero-diagonal constraint handling for allocation (TM/AP) problems.

rclsp 0.2.0 (2025-11-26)

Bug fixes

• Removed duplicated warnings.