snaykey commited on
Commit
3bcae11
·
verified ·
1 Parent(s): 7f62874

logbook.json (verbatim titles)

Browse files
Files changed (1) hide show
  1. logbook.json +23 -0
logbook.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "title": "Repro: PINE — Pruning Boosted Tree Ensembles with Conformal Equivalence",
4
+ "emoji": "microscope",
5
+ "space_id": "snaykey/repro-pine-conformal-pruning",
6
+ "paper": {"openreview_id": "LxAy8UpVUC"},
7
+ "tags": ["icml2026-repro", "paper-LxAy8UpVUC"],
8
+ "updated_at": "2026-07-27T00:00:00+00:00",
9
+ "root": {
10
+ "slug": "index",
11
+ "title": "Repro: PINE — Pruning Boosted Tree Ensembles with Conformal Equivalence",
12
+ "file": "pages/index.md",
13
+ "children": [
14
+ {"slug": "executive-summary", "title": "Executive summary", "file": "pages/executive-summary/page.md", "children": []},
15
+ {"slug": "claim-1", "title": "PINE guarantees prediction equivalence only within an in-distribution region calibrated via split conformal prediction, controlled by a single hyperparameter alpha, providing a lower bound of 1-alpha on the probability that a future input falls within the equivalence guarantee region.", "file": "pages/claim-1/page.md", "children": []},
16
+ {"slug": "claim-2", "title": "As alpha varies from 0.05 to 0.8, PINE's mean pruning rate increases from 44.6% to 67.8% while mean fidelity remains between 99.15% and 99.96% across datasets.", "file": "pages/claim-2/page.md", "children": []},
17
+ {"slug": "claim-3", "title": "PINE achieves up to a 30% improvement in compression ratio over the faithful pruning baseline FIPE while maintaining comparable prediction equivalence.", "file": "pages/claim-3/page.md", "children": []},
18
+ {"slug": "claim-4", "title": "On the Pima-Diabetes dataset at alpha=0.8, PINE attains a 55.3% pruning rate with 98.3% fidelity, versus FIPE's 17.3% pruning rate at 100% fidelity.", "file": "pages/claim-4/page.md", "children": []},
19
+ {"slug": "claim-5", "title": "PINE's plausible score function uses Chow-Liu trees to model the input distribution, enabling a compact MILP encoding with O(p*B^2) constraints rather than exponential complexity.", "file": "pages/claim-5/page.md", "children": []},
20
+ {"slug": "conclusion", "title": "Conclusion", "file": "pages/conclusion/page.md", "children": []}
21
+ ]
22
+ }
23
+ }