Run tiny-hypergraph region cost optimizer - #2152
Conversation
|
/benchmark --same-machine --dataset 18 |
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Benchmark This PRRun benchmarks by commenting on this PR: Comment Everything after Use Any PR whose title contains |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 1 improved, 1 regressed. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Changed outcomes (2)
Workflow: View run |
|
/benchmark --same-machine --dataset 18 |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 1 improved, 1 regressed. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Changed outcomes (2)
Workflow: View run |
|
/benchmark-long --dataset 18 --sample-timeout 7200s |
18 ResultsMain vs PRDataset: srj18 · Scenarios: 16 · Effort: 1x effort
Timing percentiles include solved and timed-out samples. Negative timing changes are faster. Previous main run details
PR run details
Workflow: View run |
|
/benchmark-long --dataset 18 --sample-timeout 7200s |
18 ResultsMain vs PRDataset: srj18 · Scenarios: 16 · Effort: 1x effort
Timing percentiles include solved and timed-out samples. Negative timing changes are faster. Previous main run details
PR run details
Workflow: View run |
|
/benchmark-long --dataset 18 --sample-timeout 7200s |
18 ResultsMain vs PRDataset: srj18 · Scenarios: 16 · Effort: 1x effort
Timing percentiles include solved and timed-out samples. Negative timing changes are faster. Previous main run details
PR run details
Workflow: View run |
|
/benchmark-long --dataset 18 --sample-timeout 7200s |
18 FailedBenchmark workflow ended with cancelled before completion. This can happen during setup, dependency installation, or benchmark execution. Main vs PRWorkflow: View run |
|
Fast failure-risk proxy result for 7fc78e6: The selector now uses Validation before this run: 536 routes with leave-one-dataset-out calibration gave AUC 0.818, Brier 0.153, and log loss 0.455 for predicting >100k high-density iterations. Ranking on held-out srj18 was AUC 0.881. Same-machine dataset 18 result: completion 87.5% -> 93.8%, relaxed DRC 43.8% -> 50.0%, no timeouts. The 10% burden-reduction gate selected optimized candidates only for samples 2, 6, 10, 13, and 15; it rejected the marginal changes for 3, 4, 9, 12, and 16. On samples solved by both main and PR, DRC issues fell 268 -> 196 (-26.9%). Total issues rose 284 -> 347 because newly completed samples 14 and 15 contribute 151 issues that were previously unmeasurable failures. Tradeoff: this improves completion and DRC outcomes, but it is not a speed win overall: P50 +13.3%, P60 +56.5%, P95 +23.5%. The repeated run ranked hard-search cases at AUC 0.917, though that repeat is not independent of the training corpus. Benchmark: #2152 (comment) |
|
/usf |
📸 Update Snapshots (focused)✅ Snapshot updates committed to Focused test files:
Source failure run: https://github.com/tscircuit/tscircuit-autorouter/actions/runs/32327533297 Changed snapshots:
Environment: benchmark runner ( 🔗 Workflow: View run |
|
/usf --ubuntu-latest |
📸 Update Snapshots (focused)✅ No snapshot changes were produced; nothing to commit. Focused test files:
Source failure run: https://github.com/tscircuit/tscircuit-autorouter/actions/runs/32327533297 Environment: ubuntu-latest ( 🔗 Workflow: View run |
|
This PR has been automatically marked as stale because it has had no recent activity. It will be closed if no further activity occurs. |
|
This PR was closed because it has been inactive for 1 day since being marked as stale. |
…hypergraph-166-region-cost # Conflicts: # lib/solvers/PortPointPathingSolver/tinyhypergraph/TinyHypergraphPortPointPathingSolver.ts # package.json # tests/__snapshots__/autorouting-dataset18-sample10-exact-drc-residue-visual-linux.snap.svg
|
/benchmark-long --same-machine --pipeline 7 --dataset 18 --sample-timeout 7200s |
|
/benchmark-long --same-machine --pipeline 9 --dataset 18 --sample-timeout 7200s |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 0 improved, 0 regressed. DRC issues are totaled across solved samples. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Workflow: View run |
Same Machine Benchmark ResultsBoth revisions ran sequentially in one Blacksmith job on Dataset:
Outcome changes: 0 improved, 0 regressed. DRC issues are totaled across solved samples. Timing percentiles include solved and timed-out samples; negative timing deltas are faster. Workflow: View run |
|
This PR has been automatically marked as stale because it has had no recent activity. It will be closed if no further activity occurs. |
|
This PR was closed because it has been inactive for 1 day since being marked as stale. |
3 participants
Summary
85733078optimizeRegionCostsstage for downstream routing and visualizationValidation
bun test tests/tinyhypergraph-post-solve-region-optimizer.test.ts tests/tinyhypergraph-terminal-port-ids.test.ts tests/tinyhypergraph-pipeline-error-propagation.test.ts tests/solvers/tinyhypergraph-candidate-portfolio.test.ts tests/features/partial-rip-preloaded-trace-occupancy.test.ts tests/features/tinyhypergraph-port-bridge-repro.test.ts tests/pipeline-immutability/autorouting-pipeline4-tiny-hypergraph.test.ts tests/features/pipeline7-full-pipeline-svg-frames.test.ts tests/e2e3-multisection.test.ts --timeout 9999999— 12 passedbunx tsc --noEmitbun run buildgit diff --checkRelated
main; earlier experiments: Run hypergraph-native region optimization after every solve #2115 and Use isolated post-solve unravel step in Pipeline7 #2126