jgreathouse9/mlsynth
This is the repository for the Python library mlsynth
causal-inferenceeconometricssynthetic-controlsynthetic-control-method
First Claude commit: Jun 7, 2026Last Claude commit: 17d agoDiscovered: Jul 13, 2026
Recent Claude Commits
Fix canonical band normalizer for full-length EIV/conformal bands
f3a5de817d agoauthor_emailbenchmarks: differential cross-validation of SpSyDiD on the real Arizona LAWA CPS panel
a75b51a18d agoauthor_emailAdd SCD (Synthetic Control with Differencing) estimator
c476c8618d agoauthor_emailproximal: solve cPI simplex bridge with the pure-NumPy active-set solver
18c0c4419d agoauthor_emailproximal: add cPI (simplex-constrained PIOID) + benchmark it on German reunification
1a4a83419d agoauthor_emailbenchmarks: cross-validate PROXIMAL PIOID vs the authors' manuscript code (German reunification)
f4e3b1419d agoauthor_emailproximal: PIOID ATT SE reproduces the manuscript's GMM CI (Newey-West lag 10)
8ff7d8b19d agoauthor_emailproximal: add over-identified unit-instrument PI method (PIOID)
e27fc1219d agoauthor_emailspsydid: fold the directly-treated count into the unit-weight regularization
a250e6919d agoauthor_emailsdid: fold the treated-unit count into the unit-weight regularization
8f13ad719d agoauthor_emailsdid: add jackknife and bootstrap variance estimators (vce option)
23c762019d agoauthor_emailsdid: exclude the pseudo-treated unit from its own donor pool in placebo inference
858594619d agoauthor_emailbenchmarks: compare uncertainty statistics, not just point estimates (SPSC)
2704a9319d agoauthor_emailbenchmarks: capture the doubly-robust proximal (DR-OID) cross-validation
ccae2ee19d agoauthor_emailbenchmarks: surface SPSC (single-proxy SC) in the Validation dashboard
a920f9219d agoauthor_emailbenchmarks: capture 4 missing cross-validations; surface pending ones
d28036319d agoauthor_emaildocs: public Validation dashboard; retire the comparison workbook
65060fe19d agoauthor_emailbenchmarks: add validation-dashboard generator (prototype)
c8e276219d agoauthor_emailbench(geolift_augsynth_ref): gate on the captured augsynth reference, not a live re-run
fe8b04227d agoauthor_emailRun benchmark validation at 02:00 EST (07:00 UTC) daily
7fb52051mo agoauthor_emailAdd daily benchmark validation workflow
d7398541mo agoauthor_emailMSQRT: size-weighted aggregate ATT with valid SCPI intervals
44d42461mo agoauthor_emailAdd MAREX to compare_methods; route MAREX ingestion through geoex_dataprep
86892b01mo agoauthor_emailSYNDES: holdout-validate the solution pool by default
2bbfe351mo agoauthor_emailSYNDES + MAREX: rank recommendation fit by the held-out (blank/holdout) RMSE
e6791a31mo agoauthor_emailMAREX: solution pool + MDE power + composite recommendation (SYNDES parity)
a7de2f01mo agoauthor_emailLEXSCM: expose full-precision control weights; from_lexscm uses them
31986091mo agoauthor_emailAdd LEXSCM as a third method in compare_methods
3f588781mo agoauthor_emailfrom_lexscm: use full-precision treated weights
d35cf611mo agoauthor_emailMerge remote-tracking branch 'origin/main' into claude/paper-jss-trim
cb0e6f41mo agoauthor_emailmarex_walmart: full-panel live R cross-validation (no Gurobi); fix T0 window
d4c7b471mo agoauthor_emailpaper: swap the design exemplar to MAREX on real Walmart data (placebo)
ac622cb1mo agoauthor_emailbenchmarks: SPILLSYNTH-SAR Sudan cross-validation vs the authors' Rcpp sampler
6116f901mo agoauthor_emaildocs(spillsynth): correct the Sudan SAR rho verification claim
5e310b51mo agoauthor_emailpaper: replace spillovers section with the Sudan-secession SAR illustration
13af3031mo agoauthor_emailpaper: corrected SPSC figure + hide plotting boilerplate (on trimmed base)
c079ae11mo agoauthor_emailpaper: trim MAREX and SPOTSYNTH exposition
165020e1mo agoauthor_emailpaper: tighten prose to the accepted JSS register
de6cb0a1mo agoauthor_emaildocs: cross-link Truncated History from VanillaSC/SDID placebo sections + choose tree
73f926e1mo agoauthor_emailfeat: add Truncated History (TH) robustness diagnostic
0ae03741mo agoauthor_emailci: raise build/pyversions job timeout 25 -> 40 minutes
95b6a201mo agoauthor_emailMerge main into gmmsce branch
164dd4d1mo agoauthor_emailGMM-SCE: durable benchmark vs Fry's GMM-SCE.R + docs
90f2ad71mo agoauthor_emailORTHSC: add GMM-SCE (Fry 2024) as a dispatched method
cb6583d1mo agoauthor_emailbenchmarks: cross-validate VanillaSC against the original R Synth solver
c4f9a901mo agoauthor_emailpaper(jss): tighten the unified-view and related-software sections
db64b881mo agoauthor_emailMerge paper revisions: consolidate repetition, reproducibility, result schema
5a945fe1mo agoauthor_emailpaper(jss): don't claim every estimator is benchmarked
9b363d81mo agoauthor_emailpaper(jss): consolidate repetition, add reproducibility + result schema
ba75cea1mo agoauthor_email