Optimize CallawaySantAnna covariate path (5-19x speedup)#194

Open

igerber wants to merge 25 commits intomainfrom

speed-review

Commits on Mar 14, 2026

Address PR #194 review round 4: NaN-cell preservation, empty covariates, batch warnings, validator fields

igerber
and
claude
committed
Filter NaN cells in bootstrap aggregation paths, add bootstrap NaN test, extend validator group fields

igerber
and
claude
committed
Fix overly strict AI PR review prompt causing excessive review cycles

igerber
and
claude
committed
Harden AI review against spoofed prior reviews and TODO.md loopholes

igerber
and
claude
committed
Document exact AI reviewer deviation label formats in dev docs

igerber
and
claude
committed
Fix deviation label spelling: move colon inside bold markers

igerber
and
claude
committed
Add EfficientDiD estimator (Chen, Sant'Anna & Xie 2025, Phase 1)

igerber
and
claude
committed
Address PR #192 review feedback: fix PT-Post baseline, add WIF correction, harden validation

igerber
and
claude
committed
Fix PT-All index set, bootstrap NaN filtering, and cohort drop warning

igerber
and
claude
committed
Address PR #192 review (Round 3): add param validation, fix WIF test, document conventions

igerber
and
claude
committed
Address PR #192 review (Round 4): fix balance_e NaN mismatch, reject duplicate rows, strengthen docs

igerber
and
claude
committed
Address PR #192 review (Round 5): balance_e empty warning, inline comments for repeat false positives

igerber
and
claude
committed
Address PR #192 review (Round 7): fix REGISTRY labels, remove unused params, add inline comments

igerber
and
claude
committed
Fix TROP nuclear norm solver: correct proximal gradient step size + FISTA acceleration

igerber
and
claude
committed
Fix twostep nuclear norm solver threshold scaling for non-uniform weights

igerber
and
claude
committed
Guard W_max==0 division in twostep nuclear norm solver + update docstrings

igerber
and
claude
committed
Add (1-W) weight masking to TROP global method + rename joint→global

igerber
and
claude
committed
Fix stale coefficients in global low-rank solver and NaN bootstrap poisoning

igerber
and
claude
committed
Fix Rust convergence criterion, n_valid_treated consistency, and NaN bootstrap SE

igerber
and
claude
committed
Fix global bootstrap NaN SE test to disable Rust backend

igerber
and
claude
committed
Use filtered post_treatment_indices for original_overall, fix test to poison post-treatment cell, add one-sided NaN detection in validator

igerber
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimize CallawaySantAnna covariate path (5-19x speedup)#194

Optimize CallawaySantAnna covariate path (5-19x speedup)#194
igerber wants to merge 25 commits intomainfrom
speed-review

Commits on Mar 8, 2026

Optimize CallawaySantAnna covariate path with Cholesky and pscore caching

Fix not_yet_treated cache keys and add dropped-cell warning

Restore rank-deficiency handling, complete IF schema migration, strengthen validation

Gate Cholesky behind rank checks, add reduced-column solve and cond=1e-07

Commits on Mar 14, 2026

Address PR #194 review round 4: NaN-cell preservation, empty covariates, batch warnings, validator fields

Filter NaN cells in bootstrap aggregation paths, add bootstrap NaN test, extend validator group fields

Fix overly strict AI PR review prompt causing excessive review cycles

Harden AI review against spoofed prior reviews and TODO.md loopholes

Document exact AI reviewer deviation label formats in dev docs

Fix deviation label spelling: move colon inside bold markers

Add EfficientDiD estimator (Chen, Sant'Anna & Xie 2025, Phase 1)

Address PR #192 review feedback: fix PT-Post baseline, add WIF correction, harden validation

Fix PT-All index set, bootstrap NaN filtering, and cohort drop warning

Address PR #192 review (Round 3): add param validation, fix WIF test, document conventions

Address PR #192 review (Round 4): fix balance_e NaN mismatch, reject duplicate rows, strengthen docs

Address PR #192 review (Round 5): balance_e empty warning, inline comments for repeat false positives

Address PR #192 review (Round 7): fix REGISTRY labels, remove unused params, add inline comments

Fix TROP nuclear norm solver: correct proximal gradient step size + FISTA acceleration

Fix twostep nuclear norm solver threshold scaling for non-uniform weights

Guard W_max==0 division in twostep nuclear norm solver + update docstrings

Add (1-W) weight masking to TROP global method + rename joint→global

Fix stale coefficients in global low-rank solver and NaN bootstrap poisoning

Fix Rust convergence criterion, n_valid_treated consistency, and NaN bootstrap SE

Fix global bootstrap NaN SE test to disable Rust backend

Use filtered post_treatment_indices for original_overall, fix test to poison post-treatment cell, add one-sided NaN detection in validator