Conversation
- Enhanced SDPT3 termcode mapping with more precise status categories - Added STALLED status for algorithm progress stagnation (termcodes -2, -5, -7, -8, -9) - Added NUM_ERROR for true numerical failures (termcodes 3, -3, -4) - Improved MAX_ITER handling for iteration limits (termcodes -1, -6) - Updated HTML report generator with new status styles and legend - Better distinction between numerical errors and algorithmic stagnation 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
- Enhanced SEDUMI numerr code mapping with precision distinction - Added 'optimal (inaccurate)' status for numerr=1 (matches CVXPY convention) - Kept num_error status for numerr=2 (complete numerical failure) - Updated HTML report generator to handle 'optimal (inaccurate)' consistently - Better diagnostic capability for SEDUMI numerical issues Changes: - SEDUMI numerr=1: 'optimal (inaccurate)' (solution found but low accuracy) - SEDUMI numerr=2: 'num_error' (complete numerical failure) - Status format now consistent with CVXPY: 'OPTIMAL (INACCURATE)' - Removed incorrect 'optimal_inaccurate' format from HTML report 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
- Added original_status field to SDPT3 and SEDUMI result structures - Preserved complete solver-specific status information in database - Enhanced reproducibility and debugging capabilities for research use SDPT3 original status preservation: - termcode: SDPT3's internal termination code (-2, -3, etc.) - iter: iteration count - full_info: complete info structure from SDPT3 SEDUMI original status preservation: - pinf: primal infeasibility flag (0/1) - dinf: dual infeasibility flag (0/1) - numerr: numerical error level (0/1/2) - feasratio: feasibility indicator - full_info: complete info structure from SEDUMI Implementation: - Modified MATLAB runner functions to include original_status - Updated JSON conversion pipeline to preserve original status - Extended Python interface to store original_status in additional_info - Verified database storage contains complete solver metadata Benefits: - Future status mapping changes can reinterpret historical data - Complete traceability for debugging solver behavior - Research publications can reference exact solver exit codes - Enhanced reproducibility for scientific use 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
- Extended original_status preservation to Python solver ecosystem
- Enhanced reproducibility and debugging for CVXPY and SciPy solvers
- Comprehensive solver metadata collection for research applications
CVXPY original status preservation:
- cvxpy_status: Original CVXPY status string ('optimal', 'optimal_inaccurate', etc.)
- cvxpy_value: Objective function value from CVXPY
- solver_stats: Complete solver statistics (solve_time, num_iters, extra_stats)
SciPy original status preservation:
- scipy_status: Original SciPy status code (0=optimal, 1=max_iter, etc.)
- scipy_success: Success boolean flag
- scipy_message: Detailed solver message (includes backend info)
- scipy_fun: Objective function value
- scipy_nit: Number of iterations
- Backend-specific metrics: crossover_nit, mip_dual_bound, mip_gap, etc.
Implementation:
- Added original_status to additional_info in both CVXPY and SciPy runners
- Preserved all available solver-specific metadata
- Maintained backward compatibility with existing result format
- Verified database storage contains complete original information
Testing verified:
- CVXPY CLARABEL: cvxpy_status='optimal_inaccurate', complete solver_stats
- SciPy HiGHS: scipy_status=0, scipy_success=True, detailed backend info
Benefits:
- Complete traceability for all solver backends
- Enhanced debugging capabilities for numerical issues
- Research-grade reproducibility for academic publications
- Future-proof result reinterpretation capabilities
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add comprehensive status code documentation to detail_design.md - Document NUM_ERROR, STALLED, MAX_ITER, OPTIMAL (INACCURATE) status codes - Add complete original_status structure for MATLAB and Python solvers - Update basic_design.md with reproducibility enhancement mention - Ensure documentation consistency with new original status feature 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
- Add Netlib repository as git submodule (94 LP problems) - Implement MPS file loader with SeDuMi format conversion - Register MPS loader in problem_interface.py - Add all 94 Netlib problems to problem_registry.yaml - Convert inequality constraints to equality with slack variables - Handle variable bounds through cone structure https://claude.ai/code/session_014WRtEFZErWR9kZd5dXoHTq
- Remove unused imports (List, Tuple, Optional, csr_matrix) - Fix import order to comply with isort rules - Rename unused loop variables to _lo, _up convention https://claude.ai/code/session_014WRtEFZErWR9kZd5dXoHTq
- Add NETLIB to KNOWN_LIBRARIES in registry integrity test - Fix test_problem_files_exist skip condition to require all submodules - Add mps to expected supported_formats in statistics test https://claude.ai/code/session_014WRtEFZErWR9kZd5dXoHTq
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014WRtEFZErWR9kZd5dXoHTq
…P' into develop # Conflicts: # docs/development/detail_design.md # scripts/reporting/html_generator.py # scripts/solvers/matlab/matlab_process_interface.py # scripts/solvers/python/cvxpy_runner.py # scripts/solvers/python/scipy_runner.py
Post-merge cleanup: ruff format normalized the manually re-applied original_status blocks in cvxpy_runner.py and scipy_runner.py.
…t modules The html_generator.py -> report module split (from the netlib branch merge) dropped the STALLED/MAX_ITER/NUM_ERROR CSS classes and status legend entries that develop had added to the old monolithic html_generator.py. Re-add them to results_matrix_report.py and raw_data_report.py, and extend raw_data_report.py's status_class mapping to also cover TIMEOUT/SIGKILL/SUBPROCESS_ERROR/UNKNOWN (previously only handled in the results matrix view).
Run the 94-problem NETLIB LP library against all 9 registered Python solvers (846 runs, 100% completion rate: 594 OPTIMAL, 42 OPTIMAL (INACCURATE), 121 ERROR, 31 UNBOUNDED, 28 INFEASIBLE, 27 SUBPROCESS_ERROR, 3 TIMEOUT). Regenerated HTML reports and JSON/CSV data exports to include the new results.
Port the NETLIB MPS parsing/conversion logic (scripts/data_loaders/python/mps_loader.py) to MATLAB as scripts/data_loaders/matlab/mps_loader.m, and wire it into matlab_solver_runner.m for file_type='mps'. This lets matlab_sedumi and matlab_sdpt3 run NETLIB problems, which previously failed with "Unsupported file type: mps". Unlike the Python loader, variables are explicitly reordered so free variables (K.f) come before nonnegative variables (K.l), matching SeDuMi's strict cone ordering convention. Validated against afiro (no free vars, matches known optimal -464.7531) and capri (14 free vars, matches the known optimal 2690.0129 exactly).
Run the 94-problem NETLIB LP library against matlab_sedumi and matlab_sdpt3 (188 runs, 100% completion rate). matlab_sedumi: 83 OPTIMAL, 5 OPTIMAL (INACCURATE), 4 SUBPROCESS_ERROR, 1 each of UNBOUNDED/TIMEOUT/NUM_ERROR/INFEASIBLE. matlab_sdpt3: 59 OPTIMAL, 17 MAX_ITER, 13 STALLED, 4 SUBPROCESS_ERROR, 1 each of TIMEOUT/INFEASIBLE/ERROR. Regenerated HTML reports and JSON/CSV data exports to include the new results.
Populate known_objective_value for 93 of the 94 NETLIB registry entries from the official NETLIB LP problem summary table (netlib.org/lp/data/readme, MINOS 5.3 on VAX). standgub is left without a value per the source's own caveat (MINOS cannot report an optimal value for its GUB markers). Add docs/guides/NETLIB_KNOWN_OBJECTIVES.md documenting the source, its caveats, the registry-key-to-NETLIB-name mapping, and a data quality issue found while cross-checking: mps_loader.py's mishandling of free-variable ordering causes wrong OPTIMAL objective values for CVXPY/SciPy backends on the 11 NETLIB problems with FR/MI bounds (not fixed here — tracked as a known issue).
Reflect the newly populated known_objective_value fields in the accuracy comparisons shown in results_matrix.html/raw_data.html and the JSON/CSV data exports.
📊 Preview Reports Ready!Your preview has been deployed and is available at: Preview Details
Available Reports
Notes
Preview deployment powered by GitHub Actions |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This brings
mainup to date withdevelop, which had diverged significantly. Highlights from this session's work:claude/add-netlib-benchmark-aCnsPbranch (94 NETLIB problems, MPS loader, registry entries) and resolved conflicts with develop's original-status-preservation work.matlab_sedumi/matlab_sdpt3(188 runs).scripts/data_loaders/matlab/mps_loader.m): NETLIB problems previously failed on MATLAB solvers ("Unsupported file type: mps"); this adds proper support, including correct SeDuMi free/nonnegative variable ordering (validated against known optimal values).known_objective_valuefor 93/94 NETLIB problems from the official NETLIB LP summary table (netlib.org/lp/data/readme), documented indocs/guides/NETLIB_KNOWN_OBJECTIVES.md.scripts/data_loaders/python/mps_loader.pydoesn't reorder free variables, causing wrongOPTIMALobjective values from CVXPY/SciPy backends on 11 NETLIB problems withFR/MIbounds.develop-only history (subprocess isolation architecture, timeout controls, Docker attempt + revert, database management refactors, report redesigns, etc.) that had not yet been merged tomain.Test plan
pytest tests/passes (75 passed, 1 skipped — submodule-dependent)ruff check/ruff format --checkpasspython main.py --validatepasses (9/9 Python solvers)