SPB Git
spboucher.ai

spb/wp28_uqo

Public

WP28: table generator, PDF validation, boundary finite-m prediction

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Simon-Pierre Boucher committed (Oct 28, 2026) parent b9e37e8 Browse files

54 changed files +19,970 −2

54 changed files
Load all 54 files Unified Split
added experiments/EXPERIMENT_MANIFEST.csv +50 −0
@@ -0,0 +1,50 @@
1+run_id,git_commit,python_version,numpy,dependency_lock_sha256,seed,N,K,m,p,DGP,method,optimizer,tolerance,replicates,parallel_backend,node_count,hardware,OS,wall_seconds,communicated_bytes,rounds,status,raw_output_sha256
2+exp_contam_p1_N1048576_K65536,,3.14.5,2.5.4,,627279717397746692,1048576,65536,16,1,"{""kind"": ""contaminated"", ""eps"": 0.01, ""scale"": 10.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,471.5,0,in-memory (no network),completed,46a6acf536f806703d68926b720fa1e000170c4a6960d67973b87a884e065371
3+exp_contam_p1_N262144_K16384,,3.14.4,2.5.4,,7759161092581730262,262144,16384,16,1,"{""kind"": ""contaminated"", ""eps"": 0.01, ""scale"": 10.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre6 arm64,macOS,102.6,0,in-memory (no network),completed,ff71a3b56bc44c884081dab688c814c46e4ebcbb7e405fd66883fe7b2fc51462
4+exp_contam_p1_N262144_K4096,,3.14.4,2.5.4,,7503374439635565362,262144,4096,64,1,"{""kind"": ""contaminated"", ""eps"": 0.01, ""scale"": 10.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre4 arm64,macOS,30.0,0,in-memory (no network),completed,5763e1de245ee5cc2a696881284de4cc412508c31ca34fb6982d6195a27f33a3
5+exp_p1_N16384_K1024,7a86c23,3.14.4,2.4.4,,3432885715831462714,16384,1024,16,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,13.6,0,in-memory (no network),completed,ac691e3e603663fd952ba0c28b3870437d8dfd827c29da1b2924679d7f5d5a85
6+exp_p1_N16384_K128,7a86c23,3.14.4,2.4.4,,5777488189351826881,16384,128,128,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,3.1,0,in-memory (no network),completed,661e9e7bd93fd43c606ea8fee40fca8bbc15ad29f78e8af4ffb56a0075612387
7+exp_p1_N16384_K16,7a86c23,3.14.4,2.4.4,,5839257743945981181,16384,16,1024,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,1.8,0,in-memory (no network),completed,c74343c729af35981082f5ea96355d21b38ab0149c1bdb007cb7036407948b3c
8+exp_p1_N16384_K2048,7a86c23,3.14.4,2.4.4,,2396459402953418584,16384,2048,8,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,25.7,0,in-memory (no network),completed,16f5ed011a82d3312cb7b1c1aad2de4d2600fde23c986bb2b3967a0d5889d95a
9+exp_p1_N16384_K256,7a86c23,3.14.4,2.4.4,,3820358254920617216,16384,256,64,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,4.6,0,in-memory (no network),completed,8b2de298223c8dff1b32ee92e236b8d57a266dde51af8bf341c3bd2e35ed726a
10+exp_p1_N16384_K32,7a86c23,3.14.4,2.4.4,,323847821137284353,16384,32,512,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,2.0,0,in-memory (no network),completed,f7a0dce586d8595c68049f9faadcac7269e6413d5a997f208fce9c3224f7fdf6
11+exp_p1_N16384_K512,7a86c23,3.14.4,2.4.4,,5425295811703503756,16384,512,32,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,7.6,0,in-memory (no network),completed,dd041efa277e616d84abde21bd61623fd6ecf3758b070a210967bf26c6f34b0f
12+exp_p1_N16384_K64,7a86c23,3.14.4,2.4.4,,2806566115357425723,16384,64,256,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,2.4,0,in-memory (no network),completed,be66e03bb1260c98f95e524b75daba7d95f849c381f0d6d09b7564f97d2bf2df
13+exp_p1_N262144_K1024,2718842,3.14.4,2.4.4,,8000149967679761745,262144,1024,256,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,21.7,0,in-memory (no network),completed,2e6d59208e7f09d086d2bda273b68548f287633bb705354046e6ed3f9f408576
14+exp_p1_N262144_K16384,63bdb07,3.14.4,2.4.4,,129339267247191325,262144,16384,16,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,210.8,0,in-memory (no network),completed,ce6f26d9e30403c971f5a857b943a3ac40b6a07c7dd1bda33c22860cf90de334
15+exp_p1_N262144_K2048,63bdb07,3.14.4,2.4.4,,281599325893617438,262144,2048,128,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,34.9,0,in-memory (no network),completed,d6623c94836af69e70fab80b5dd0a3da25b26af45c598137895f66147ce6fdd8
16+exp_p1_N262144_K256,2718842,3.14.4,2.4.4,,7315058885158037941,262144,256,1024,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,11.5,0,in-memory (no network),completed,7bb6200155c7c3df4079fe2233d69a43128d2dc49b40c3ae2e83d37b8be027aa
17+exp_p1_N262144_K32768,63bdb07,3.14.4,2.4.4,,6502579563820440675,262144,32768,8,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,406.0,0,in-memory (no network),completed,354a80b07e1c24b9d96cba97aed2ffd50f381c10d5fa192e046837cf18ac646e
18+exp_p1_N262144_K4096,63bdb07,3.14.4,2.4.4,,4842079901205516085,262144,4096,64,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,62.5,0,in-memory (no network),completed,629ce95b7b2ac6ca3ec1302ce77a3b223bce374fafbe69b52d09eb8dc3c59c33
19+exp_p1_N262144_K512,2718842,3.14.4,2.4.4,,6815573501771189831,262144,512,512,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,15.3,0,in-memory (no network),completed,59885e79e3e6349fe1a2c92854426090fd9c6c82d8422f4680b8698896c319ec
20+exp_p1_N262144_K8192,63bdb07,3.14.4,2.4.4,,4526677405690696884,262144,8192,32,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,113.9,0,in-memory (no network),completed,409e9543f73b6a17ca87fb52d8567d82cbc172ba8befe77569f1044787ab1d2d
21+exp_p1_N65536_K1024,7a86c23,3.14.4,2.4.4,,3718237952087232854,65536,1024,64,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,15.3,0,in-memory (no network),completed,46077d73a20c1aac90e8d8687a688380e805c630c7b735bf134c66681df77732
22+exp_p1_N65536_K128,7a86c23,3.14.4,2.4.4,,5161033798656750738,65536,128,512,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,4.1,0,in-memory (no network),completed,47f73dc6dbf3f2e2a68cb6e86664c25eb178cce79da1ed569fb627bb185d4f6c
23+exp_p1_N65536_K2048,7a86c23,3.14.4,2.4.4,,6516373625071197227,65536,2048,32,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,27.2,0,in-memory (no network),completed,3cfe02dcd5ee7128c165d176346c0dce3103e95b4c372f1ee87d5c8443304363
24+exp_p1_N65536_K256,7a86c23,3.14.4,2.4.4,,2989917426596495497,65536,256,256,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,5.7,0,in-memory (no network),completed,3f9d416d2fe15e08cda72f02e937a4087f895272d991c6143724df1d8f2ebf03
25+exp_p1_N65536_K4096,7a86c23,3.14.4,2.4.4,,1098385355741143893,65536,4096,16,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,53.1,0,in-memory (no network),completed,5e162e892676f9b4bca11f8258a21d3366939491d76dc27a9edc13edf535e151
26+exp_p1_N65536_K512,7a86c23,3.14.4,2.4.4,,7735086372791956174,65536,512,128,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,8.9,0,in-memory (no network),completed,f51dfc54a45c13061eba7dc65cc8547fad5d3b7c19589e3cadae4c5fd70d43d9
27+exp_p1_N65536_K64,7a86c23,3.14.4,2.4.4,,7047651843593800474,65536,64,1024,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,3.3,0,in-memory (no network),completed,7f6546e60349a2d4d1c661456b5ef6f852cecdcd0bde5281b0ffd4d7297af3a1
28+exp_p1_N65536_K8192,7a86c23,3.14.4,2.4.4,,25267133336224275,65536,8192,8,1,"{""theta0"": 1.0}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,2000,xargs -P / nohup,1,MacBookPro-de-simon-pierre-2.local arm64,macOS,99.0,0,in-memory (no network),completed,5cdc53c036f5807f0f15e7ce3c56c5b98211d8a93dd3089f93b05ae08e83abb1
29+logit_misspec_p3_N262144_K1024,,3.14.4,2.5.4,,5430299985042827137,262144,1024,256,3,"{""kind"": ""misspecified"", ""omitted"": 0.5}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre4 arm64,macOS,807.5,0,in-memory (no network),completed,4ba4f64a30ce7d9379a36dcbd72dd956184887372ca115da268192acb6106b90
30+logit_misspec_p3_N262144_K2048,,3.14.4,2.5.4,,3365958059589350090,262144,2048,128,3,"{""kind"": ""misspecified"", ""omitted"": 0.5}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre6 arm64,macOS,1021.5,0,in-memory (no network),completed,0901fdcc31ac28945fc62be6fabc8bbd5906aab36a6367fabb58e0b5fb489d25
31+logit_misspec_p3_N65536_K128,,3.9.6,2.0.2,,1189149877376627910,65536,128,512,3,"{""kind"": ""misspecified"", ""omitted"": 0.5}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudonpierre arm64,macOS,261.1,0,in-memory (no network),completed,d5549965eb18106d51676e09521787a845070e08dbdceda307a18283dfd7c769
32+logit_misspec_p3_N65536_K512,,3.14.5,2.5.4,,3578458641162309222,65536,512,128,3,"{""kind"": ""misspecified"", ""omitted"": 0.5}",estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,283.3,0,in-memory (no network),completed,c29ffec0dbff1a6717c394274a6da063592ec08ce27125dd8cdad93c01dcb07b
33+logit_p3_N16384_K16,,3.14.5,2.5.4,,8206780370864179533,16384,16,1024,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,50.3,0,in-memory (no network),completed,4cc10b3c620eda3c0739cd1df61d79e3fafc59a4c7f6644ff66af9ce00acda34
34+logit_p3_N262144_K1024,,3.14.4,2.5.4,,3714220950796891036,262144,1024,256,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre4 arm64,macOS,804.2,0,in-memory (no network),completed,a5c92688046ea20d4ecaa729c03ff3a11141184c01d147551f2837cc9e73e678
35+logit_p3_N262144_K2048,,3.14.4,2.5.4,,7754184961919789368,262144,2048,128,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre6 arm64,macOS,1067.9,0,in-memory (no network),completed,1cd8e8b90ae76df5e4cdbf42222c7a23ce4944e4b957fccb4d59f97f9c0c4d97
36+logit_p3_N65536_K128,,3.9.6,2.0.2,,53022686123595153,65536,128,512,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudonpierre arm64,macOS,265.6,0,in-memory (no network),completed,9b75a4ef55ef11638fbf0c92d2274240a67b9e17a2d649f0fca6e36590f750da
37+logit_p3_N65536_K64,,3.14.4,2.5.4,,6021038287837971332,65536,64,1024,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre6 arm64,macOS,165.1,0,in-memory (no network),completed,a8f336272c233ccb70c8a75454ed30d1288f67b638b69dd2dd27e88a8e0f415d
38+logit_p8_N16384_K64,,3.14.5,2.5.4,,1259612092362166754,16384,64,256,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,117.9,0,in-memory (no network),completed,7c4cfadc511b65c4e7a548b7df23ed58e17ac3857e41c0deaef57ec4fb69f233
39+pois_p3_N16384_K16,,3.14.4,2.5.4,,1829993567022379337,16384,16,1024,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre4 arm64,macOS,34.7,0,in-memory (no network),completed,a6cc84e25fff7fe3eb398fb8e4896dd4a216a8ea9916f3a622ecda9a00f81858
40+pois_p3_N16384_K32,,3.14.4,2.5.4,,7075387799206061791,16384,32,512,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre6 arm64,macOS,35.1,0,in-memory (no network),completed,a76144b0f05d829cd82f4447059c41cf111e76af77fb433c26caf7ec4445ced9
41+pois_p3_N16384_K64,,3.14.5,2.5.4,,2295780102078045441,16384,64,256,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,44.9,0,in-memory (no network),completed,797b95205f6733adaede50defbe717619b78d518171fe4a6a42155e6b1998e71
42+pois_p3_N65536_K1024,,3.14.5,2.5.4,,198767958255358910,65536,1024,64,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,201.2,0,in-memory (no network),completed,f0ae7fa4e9b5118ed84e3900845cf1991dcc1e43f93a00dc2d18e03b12b03876
43+pois_p3_N65536_K2048,,3.14.4,2.5.4,,7728696182291572316,65536,2048,32,3,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre4 arm64,macOS,303.1,0,in-memory (no network),completed,dd5aa3abe2f8e2842e3589887feec479f4d7e7d923a1b00fe319d58355ff73d1
44+pois_p8_N16384_K16,,3.14.5,2.5.4,,3921391536461478375,16384,16,1024,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre2 arm64,macOS,86.3,0,in-memory (no network),completed,e81ded52f665f5e9543ad73df10017f392fcbded8ec598ba6762b7683335fabb
45+pois_p8_N262144_K1024,,3.9.6,2.0.2,,3579026518555065084,262144,1024,256,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,m6mb arm64,macOS,1451.3,0,in-memory (no network),completed,771f79745e9cb423c8374df301fe4cb0dc0e4e5dd8096f35308b56c95a9fd010
46+pois_p8_N262144_K256,,3.14.4,2.5.4,,8250868076379924612,262144,256,1024,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,m3u96b arm64,macOS,1341.7,0,in-memory (no network),completed,500be0cd576cf7353f5537bd343692285ba45ec7289b08a21c72bb282c18e742
47+pois_p8_N262144_K512,,3.9.6,2.0.2,,1284184495569829260,262144,512,512,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,m5ma arm64,macOS,1451.8,0,in-memory (no network),completed,e4f7a2c1b06ebcf42ad29e59997c68e09a4d51437c0c5c96dcba207399ac5c52
48+pois_p8_N65536_K1024,,3.9.6,2.0.2,,699644534752049547,65536,1024,64,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudonpierre arm64,macOS,743.8,0,in-memory (no network),completed,be439eda53ab7da32ada41fd2fbbb996b2621715d6c07d48d78f9a81fc98c828
49+pois_p8_N65536_K256,,3.14.4,2.5.4,,4910276943878145592,65536,256,256,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre4 arm64,macOS,329.6,0,in-memory (no network),completed,fb3a593c1e21575129e54df1f4e824b94e42cbefad637654da31db4cc0980bdd
50+pois_p8_N65536_K512,,3.14.4,2.5.4,,806863020632972821,65536,512,128,8,{},estimate_all,"batched Newton, step halving, tol 1e-9 (central 1e-12)",1e-9,1000,xargs -P / nohup,1,MacStudnpierre6 arm64,macOS,351.5,0,in-memory (no network),completed,be094bde5ef0e9bf2a61249827bcfb6088de0ecdcf390292ad45b889d3cac088
added experiments/boundary_pred.py +42 −0
@@ -0,0 +1,42 @@
1+"""Finite-m prediction of the mean scaled gap in the exponential boundary study.
2+E[sqrt(N)(theta^(q) - theta_hat)] ~ -sqrt(N) E[e0^(2^q)] / theta0^(2^q - 1) (q >= 1), with e0 = pilot - theta_hat,
3+E e0 = b_m (exact finite-m bias of J_r), Var e0 ~ Var(R^J_k)/K (R_N term negligible); Var(R^J_k) by exact formula (r = 1, 2) or
4+by a seeded simulation of one block (r >= 3). Gaussian moments are used for E e0^4 (q = 2)."""
5+import sys, pathlib, numpy as np
6+sys.path.insert(0, str(pathlib.Path(__file__).resolve().parents[1] / "src"))
7+from pamle.jackknife import vandermonde_weights, DEFAULT_SPLITS
8+
9+
10+def exact_bias(r, m, theta0=1.0):
11+ s = DEFAULT_SPLITS[r]; w = vandermonde_weights(s)
12+ return theta0 * sum(wj * sj / (m - sj) for wj, sj in zip(w, s))
13+
14+
15+def var_RJ(r, m, theta0=1.0, reps=400000, seed=7):
16+ if (r == 1 and m <= 2) or (r == 2 and m <= 4) or (r >= 3 and m // DEFAULT_SPLITS[r][-1] <= 2):
17+ return float('nan') # variance of the local remainder does not exist
18+ if r == 1:
19+ return theta0 ** 2 * ((3 * m + 2) / (m * (m - 1) * (m - 2)) - 1 / (m - 1) ** 2)
20+ if r == 2:
21+ varJ = theta0 ** 2 * m ** 2 * (m ** 2 - 6 * m + 17) / ((m - 4) * (m - 2) ** 2 * (m - 1) ** 2)
22+ covJU = theta0 ** 2 * (2 / (m - 1) - 1 / (m - 2)) # Cov(J2, U) with U = theta0 - theta0^2 Xbar
23+ return varJ + theta0 ** 2 / m - 2 * covJU
24+ s = DEFAULT_SPLITS[r]; w = vandermonde_weights(s); smax = s[-1]
25+ rng = np.random.default_rng(seed)
26+ parts = rng.gamma(shape=m // smax, scale=1.0 / theta0, size=(reps, smax))
27+ J = np.zeros(reps)
28+ for wj, sj in zip(w, s):
29+ sub = parts.reshape(reps, sj, smax // sj).sum(axis=2); J += wj * ((m / sj) / sub).mean(axis=1)
30+ U = theta0 - theta0 ** 2 * parts.sum(axis=1) / m
31+ return float(np.var(J - U, ddof=1))
32+
33+
34+def predicted_mean_gap(r, q, m, K, N, theta0=1.0):
35+ b = exact_bias(r, m, theta0); v = var_RJ(r, m, theta0) / K
36+ if q == 0:
37+ return np.sqrt(N) * b
38+ if q == 1:
39+ return -np.sqrt(N) * (b ** 2 + v) / theta0
40+ if q == 2:
41+ return -np.sqrt(N) * (b ** 4 + 6 * b ** 2 * v + 3 * v ** 2) / theta0 ** 3
42+ raise ValueError(q)
added experiments/make_tables.py +215 −0
@@ -0,0 +1,215 @@
1+#!/usr/bin/env python3
2+"""Generate paper/tables/*.tex (LaTeX fragments), paper/tables/numbers.tex (macros) and experiments/EXPERIMENT_MANIFEST.csv
3+from results/processed/summary.csv, results/processed/boundary_summary.csv, results/raw/ablation_*.csv, results/raw/skew_*.csv
4+and results/raw/benchmarks.jsonl. Degrades gracefully when a result set is absent (the fragment then says so)."""
5+import glob, json, pathlib, sys
6+import numpy as np, pandas as pd
7+ROOT = pathlib.Path(__file__).resolve().parents[1]
8+PROC, RAW, TAB = ROOT / "results" / "processed", ROOT / "results" / "raw", ROOT / "paper" / "tables"
9+TAB.mkdir(parents=True, exist_ok=True)
10+LAB = {"r1_q0": "(1,0) plain average", "r1_q1": "(1,1) average + 1 Newton", "r1_q2": "(1,2) average + 2 Newton", "r2_q0": "(2,0) half-block jackknife",
11+ "r2_q1": "(2,1) half-block + 1 Newton", "r2_q2": "(2,2) half-block + 2 Newton", "r3_q0": "(3,0) quarter-block", "r3_q1": "(3,1) quarter-block + 1 Newton",
12+ "r3_q2": "(3,2) quarter-block + 2 Newton", "cs_q0": "analytic Cox--Snell, $q=0$", "cs_q1": "analytic Cox--Snell + 1 Newton", "csl1": "CSL, 1 round", "csl2": "CSL, 2 rounds",
13+ "aee": "AEE (Lin--Xi)", "stale_r1_q1": "(1,1) stale Hessian", "stale_r2_q1": "(2,1) stale Hessian", "stale_r1_q2": "(1,2) stale Hessian", "central": "centralized MLE"}
14+MAIN = ["r1_q0", "r1_q1", "r1_q2", "r2_q0", "r2_q1", "r2_q2", "r3_q0", "r3_q1", "cs_q1", "csl1", "csl2", "aee", "stale_r2_q1", "central"]
15+
16+
17+def fmt(x, d=3):
18+ if x is None or (isinstance(x, float) and not np.isfinite(x)):
19+ return "--"
20+ a = abs(x)
21+ if a == 0:
22+ return "0"
23+ if a < 1e-3 or a >= 1e4:
24+ return f"{x:.2e}".replace("e-0", "e-").replace("e+0", "e+")
25+ return f"{x:.{d}f}"
26+
27+
28+def table_cell(s, label, p, N, ms, fname, caption, ests=MAIN):
29+ d = s[(s.label == label) & (s.p == p) & (s.N == N) & (s.m.isin(ms))]
30+ if d.empty:
31+ (TAB / fname).write_text(f"% no data for {label} p={p} N={N}\n"); return False
32+ ms = sorted(d.m.unique())
33+ lines = [r"\begin{table}[p]", r"\centering\scriptsize", r"\begin{threeparttable}", f"\\caption{{{caption}}}\\label{{tab:{fname[:-4]}}}",
34+ r"\begin{tabular}{@{}l" + "r" * (3 * len(ms)) + "@{}}", r"\toprule",
35+ "Estimator & " + " & ".join(f"\\multicolumn{{3}}{{c}}{{$m = {m}$, $K = {int(N//m):,}$}}" for m in ms) + r" \\",
36+ " & " + " & ".join("gap & RMSE & cov" for _ in ms) + r" \\", r"\midrule"]
37+ for e in ests:
38+ row = [LAB.get(e, e)]
39+ for m in ms:
40+ dd = d[(d.estimator == e) & (d.m == m)]
41+ if dd.empty:
42+ row += ["--", "--", "--"]; continue
43+ dd = dd.iloc[0]
44+ row += [fmt(dd["rms_gap"]), fmt(dd["rmse"]), (f"{dd['cov95']:.3f}" if "cov95" in dd and np.isfinite(dd["cov95"]) else "--")]
45+ lines.append(" & ".join(row) + r" \\")
46+ reps = int(d.reps.iloc[0]); cf = d.drop_duplicates("m")[["m", "central_fail", "valid_frac_r2", "valid_frac_r3"]] if "valid_frac_r3" in d else None
47+ lines += [r"\bottomrule", r"\end{tabular}", r"\begin{tablenotes}\footnotesize",
48+ f"\\item {reps} paired replications per column. gap $=$ root mean square of $\\sqrt N\\,\\lVert\\tilde\\theta - \\widehat\\theta_N\\rVert$; RMSE $=$ root mean square of $\\sqrt N\\,\\lVert\\tilde\\theta - \\theta_0\\rVert$; cov $=$ empirical coverage of nominal 95\\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \\texttt{{results/processed/summary.csv}}; coverage standard error $\\approx {np.sqrt(0.95*0.05/reps):.3f}$."]
49+ if cf is not None:
50+ lines.append("\\item Fraction of workers with all local fits valid (orders 2 / 3): " + "; ".join(f"$m={int(r.m)}$: {r.valid_frac_r2:.3f} / {r.valid_frac_r3:.3f}" for r in cf.itertuples()) + ".")
51+ lines += [r"\end{tablenotes}", r"\end{threeparttable}", r"\end{table}"]
52+ (TAB / fname).write_text("\n".join(lines) + "\n"); return True
53+
54+
55+def table_rates(s, fname="mc_rates.tex"):
56+ """Log-log slopes of rms_gap vs m at fixed N (largest N per model) for the main estimators."""
57+ rows = []
58+ for (label, p), d in s.groupby(["label", "p"]):
59+ N = d.N.max()
60+ for e in ["r1_q0", "r1_q1", "r1_q2", "r2_q0", "r2_q1", "r3_q0", "r3_q1", "cs_q1"]:
61+ dd = d[(d.N == N) & (d.estimator == e) & (d.rms_gap > 1e-9)].sort_values("m")
62+ if len(dd) >= 3:
63+ slope = np.polyfit(np.log(dd.m), np.log(dd.rms_gap), 1)[0]
64+ rows.append((label, p, N, e, slope, len(dd)))
65+ if not rows:
66+ (TAB / fname).write_text("% no data\n"); return
67+ th = {"r1_q0": -1, "r1_q1": -2, "r1_q2": -4, "r2_q0": -2, "r2_q1": -4, "r3_q0": -3, "r3_q1": -6, "cs_q1": -4}
68+ df = pd.DataFrame(rows, columns=["label", "p", "N", "estimator", "slope", "n"])
69+ piv = df.pivot_table(index=["label", "p", "N"], columns="estimator", values="slope")
70+ cols = [c for c in ["r1_q0", "r1_q1", "r1_q2", "r2_q0", "r2_q1", "r3_q0", "r3_q1", "cs_q1"] if c in piv.columns]
71+ lines = [r"\begin{table}[t]", r"\centering\small", r"\begin{threeparttable}", r"\caption{Empirical log-log slopes of the RMS scaled gap to $\widehat\theta_N$ against $m$ at fixed $N$ (largest $N$ of each design), with the slopes predicted by the leading bias term.}\label{tab:mc_rates}",
72+ r"\begin{tabular}{@{}llr" + "r" * len(cols) + "@{}}", r"\toprule", "Design & $p$ & $N$ & " + " & ".join(c.replace("_", "\\_") for c in cols) + r" \\",
73+ "predicted & & & " + " & ".join(str(th[c]) for c in cols) + r" \\", r"\midrule"]
74+ for idx, r in piv.iterrows():
75+ lines.append(f"{idx[0].replace('_', ' ')} & {idx[1]} & {int(idx[2]):,} & " + " & ".join(fmt(r[c], 2) for c in cols) + r" \\")
76+ lines += [r"\bottomrule", r"\end{tabular}", r"\begin{tablenotes}\footnotesize", r"\item Slopes are least-squares fits over the available $m$ grid; they approach the predicted exponent only where the bias term dominates the random remainder and the sample is in the asymptotic regime. The predicted slope for the Cox--Snell + 1 Newton estimator assumes an exact first-order bias formula ($-4$). Source: \texttt{results/processed/summary.csv}.", r"\end{tablenotes}", r"\end{threeparttable}", r"\end{table}"]
77+ (TAB / fname).write_text("\n".join(lines) + "\n")
78+
79+
80+def table_boundary(fname="boundary.tex"):
81+ f = PROC / "boundary_summary.csv"
82+ if not f.exists():
83+ (TAB / fname).write_text("% no boundary data\n"); return
84+ b = pd.read_csv(f)
85+ sys.path.insert(0, str(ROOT / "experiments")); from boundary_pred import predicted_mean_gap
86+ b["pred_finite_m"] = [predicted_mean_gap(int(r.r), int(r.q), int(r.m), int(r.K), int(r.N)) for r in b.itertuples()]
87+ b.to_csv(PROC / "boundary_summary_with_prediction.csv", index=False)
88+ lines = [r"\begin{table}[t]", r"\centering\scriptsize", r"\begin{threeparttable}", r"\caption{Boundary study in the exponential model (Theorem~\ref{thm:exponential}(v)): mean of $\sqrt N(\theta^{(q)}_r - \widehat\theta_N)$ when $K \approx cN^{\alpha(r,q)}$, against the limit $-\theta_0(\beta_r c_{\mathrm{eff}}^r)^{2^q}$ ($q\ge1$) or $\beta_r\theta_0 c_{\mathrm{eff}}^r$ ($q=0$), with $c_{\mathrm{eff}} = K/N^{\alpha}$ after rounding $m$ to a multiple of the finest split.}\label{tab:boundary}",
89+ r"\begin{tabular}{@{}ccrrrrrrrr@{}}", r"\toprule", r"$(r,q)$ & $c$ & $N$ & $K$ & $m$ & reps & mean gap (s.e.) & finite-$m$ prediction & limit at $c_{\mathrm{eff}}$ & gap/limit \\", r"\midrule"]
90+ for (r, q, c), d in b.groupby(["r", "q", "c"]):
91+ for row in d.sort_values("N").itertuples():
92+ ratio = row.mean_gap / row.limit_c_eff if row.limit_c_eff != 0 else np.nan
93+ lines.append(f"({r},{q}) & {c:g} & $2^{{{int(np.log2(row.N))}}}$ & {row.K:,} & {row.m} & {row.reps} & {row.mean_gap:+.4g} ({row.se:.2g}) & {(f'{row.pred_finite_m:+.4g}' if np.isfinite(row.pred_finite_m) else '--')} & {row.limit_c_eff:+.4g} & {ratio:.2f} \\\\")
94+ lines += [r"\bottomrule", r"\end{tabular}", r"\begin{tablenotes}\footnotesize", r"\item Exact-Gamma simulation of block sufficient statistics; seed 20261028 plus a cell hash. The ratio converges to one only as $m\to\infty$; with $m \asymp N^{1/(2^{q+1}r)}$ the local sample sizes in the table are small and the finite-$m$ bias $\theta_0\sum_j w_js_j/(m-s_j)$ exceeds its leading term. Source: \texttt{results/processed/boundary\_summary.csv}.", r"\end{tablenotes}", r"\end{threeparttable}", r"\end{table}"]
95+ (TAB / fname).write_text("\n".join(lines) + "\n")
96+
97+
98+def table_ablation(model, fname):
99+ files = sorted(glob.glob(str(RAW / f"ablation_{model}_*.csv")))
100+ if not files:
101+ (TAB / fname).write_text(f"% no ablation data for {model}\n"); return None
102+ d = pd.concat([pd.read_csv(f) for f in files], ignore_index=True)
103+ d = d[np.isfinite(d.gap)]
104+ g = d.groupby(["N", "m", "variant", "pattern", "r", "q", "hessian", "bits"]).agg(rms=("gap", lambda x: np.sqrt(np.mean(x ** 2))), mean0=("gap0", "mean"), n=("gap", "size"), valid=("valid_frac", "mean")).reset_index()
105+ g.to_csv(PROC / f"ablation_{model}_summary.csv", index=False)
106+ # (a) patterns x q at the two largest m of the largest N
107+ N = g.N.max(); ms = sorted(g[g.N == N].m.unique())
108+ lines = [r"\begin{table}[p]", r"\centering\scriptsize", r"\begin{threeparttable}", f"\\caption{{Ablation ({model}, $N = {N:,}$): RMS of $\\sqrt N\\,\\lVert\\tilde\\theta - \\widehat\\theta_N\\rVert$ by split pattern, order $r$, Newton rounds $q$ and curvature used (exact / stale averaged local / single machine); quantized exact curvature with $b$ mantissa bits for $q = 1$.}}\\label{{tab:abl_{model}}}",
109+ r"\begin{tabular}{@{}llr" + "r" * len(ms) + "@{}}", r"\toprule", "Pattern & curvature & $q$ & " + " & ".join(f"$m = {m}$" for m in ms) + r" \\", r"\midrule"]
110+ for pat in ["1", "1-2", "1-3", "1-2-4", "1-2-3", "1-2-4-8", "1-2-3-4"]:
111+ for hess in ["exact", "stale", "single"]:
112+ for q in range(0, 4):
113+ if hess != "exact" and q == 0:
114+ continue
115+ vals = []
116+ for m in ms:
117+ dd = g[(g.N == N) & (g.m == m) & (g.pattern == pat) & (g.q == q) & (g.hessian == hess) & (g.bits == 64)]
118+ vals.append(fmt(float(dd.rms.iloc[0])) if len(dd) else "--")
119+ if all(v == "--" for v in vals):
120+ continue
121+ lines.append(f"({pat.replace('-', ',')}) & {hess} & {q} & " + " & ".join(vals) + r" \\")
122+ lines.append(r"\midrule")
123+ for pat in ["1", "1-2", "1-2-4"]:
124+ for hess, lab in [("quantizedH", "$b$-bit Hessian"), ("quantizedg", "$b$-bit score"), ("quantizedBoth", "$b$-bit both")]:
125+ for bits in sorted(g.bits.unique()):
126+ if bits == 64:
127+ continue
128+ vals = []
129+ for m in ms:
130+ dd = g[(g.N == N) & (g.m == m) & (g.pattern == pat) & (g.q == 1) & (g.hessian == hess) & (g.bits == bits)]
131+ vals.append(fmt(float(dd.rms.iloc[0])) if len(dd) else "--")
132+ if all(v == "--" for v in vals):
133+ continue
134+ lines.append(f"({pat.replace('-', ',')}) & {lab}, $b={bits}$ & 1 & " + " & ".join(vals) + r" \\")
135+ reps = int(d.groupby(["N", "m"]).rep.nunique().max())
136+ lines += [r"\bottomrule", r"\end{tabular}", r"\begin{tablenotes}\footnotesize", f"\\item {reps} paired replications per column; $K = N/m$. Rows are omitted when the pattern does not divide $m$. Source: \\texttt{{results/processed/ablation\\_{model}\\_summary.csv}}.", r"\end{tablenotes}", r"\end{threeparttable}", r"\end{table}"]
137+ (TAB / fname).write_text("\n".join(lines) + "\n"); return g
138+
139+
140+def table_skew(fname="skew.tex"):
141+ files = sorted(glob.glob(str(RAW / "skew_*.csv")))
142+ if not files:
143+ (TAB / fname).write_text("% no skew data\n"); return
144+ d = pd.concat([pd.read_csv(f) for f in files], ignore_index=True)
145+ g = d.groupby(["model", "K", "ratio", "agg", "pilot"]).agg(gap0=("gap_q0", lambda x: np.sqrt(np.mean(x ** 2))), gap1=("gap_q1", lambda x: np.sqrt(np.mean(x ** 2))), gap2=("gap_q2", lambda x: np.sqrt(np.mean(x ** 2))), n=("rep", "size"), N=("N", "first")).reset_index()
146+ g.to_csv(PROC / "skew_summary.csv", index=False)
147+ lines = [r"\begin{table}[t]", r"\centering\scriptsize", r"\begin{threeparttable}", r"\caption{Unequal block sizes (Proposition~\ref{prop:robust}): RMS of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$ for size-weighted and equal-weight aggregation of the raw and half-block-jackknife local estimates, with $q = 0, 1, 2$ Newton corrections. Sizes grow geometrically with max/min ratio as shown; $N \approx 2^{18}$.}\label{tab:skew}",
148+ r"\begin{tabular}{@{}llrrllrrr@{}}", r"\toprule", r"Model & $K$ & ratio & reps & weights & pilot & $q=0$ & $q=1$ & $q=2$ \\", r"\midrule"]
149+ for r in g.sort_values(["model", "K", "ratio", "agg", "pilot"]).itertuples():
150+ lines.append(f"{r.model} & {r.K} & {r.ratio:g} & {r.n} & {r.agg} & {r.pilot} & {fmt(r.gap0)} & {fmt(r.gap1)} & {fmt(r.gap2)} \\\\")
151+ lines += [r"\bottomrule", r"\end{tabular}", r"\begin{tablenotes}\footnotesize", r"\item Source: \texttt{results/processed/skew\_summary.csv}.", r"\end{tablenotes}", r"\end{threeparttable}", r"\end{table}"]
152+ (TAB / fname).write_text("\n".join(lines) + "\n")
153+
154+
155+def table_bench(fname="bench.tex"):
156+ f = RAW / "benchmarks.jsonl"
157+ if not f.exists():
158+ (TAB / fname).write_text("% no benchmark data\n"); return None
159+ rows = [json.loads(l) for l in open(f) if l.strip()]
160+ recs = []
161+ for r in rows:
162+ st = r["stats"]; A = st["stages"]["A_B_local_fit_and_upload"]; rounds = [v for k, v in st["stages"].items() if k.startswith("D_F")]
163+ recs.append({"label": r["label"], "K": r["K"], "hosts": r["n_hosts"], "p": r["p"], "m": r["m"], "r": len(r["splits"]), "q": r["q"], "hessian": r["hessian_mode"],
164+ "t_total": r["t_total_after_connect"], "t_local": A["wall"], "fit_mean": A["worker_fit_s_mean"], "fit_max": A["worker_fit_s_max"], "lat99_A": A["latency_p99"],
165+ "t_rounds": sum(v["wall"] for v in rounds), "round_wall": np.mean([v["wall"] for v in rounds]) if rounds else 0.0, "der_max": max([v["worker_der_s_max"] for v in rounds]) if rounds else 0.0,
166+ "lat99_round": max([v["latency_p99"] for v in rounds]) if rounds else 0.0, "up_B": st["bytes_up"], "down_B": st["bytes_down"], "msgs": st["messages_up"] + st["messages_down"], "grad_final": r["newton"][-1]["grad_norm"] if r["newton"] else np.nan, "t_connect": r["t_connect"]})
167+ d = pd.DataFrame(recs)
168+ g = d.groupby(["label", "K", "hosts", "p", "r", "q", "hessian"]).agg(reps=("t_total", "size"), t_total=("t_total", "median"), t_local=("t_local", "median"), fit_mean=("fit_mean", "median"), fit_max=("fit_max", "median"),
169+ t_rounds=("t_rounds", "median"), round_wall=("round_wall", "median"), der_max=("der_max", "median"), lat99=("lat99_round", "median"), up_B=("up_B", "first"), down_B=("down_B", "first"), msgs=("msgs", "first"), grad=("grad_final", "median"), t_connect=("t_connect", "median")).reset_index()
170+ g.to_csv(PROC / "benchmarks_summary.csv", index=False)
171+ lines = [r"\begin{table}[p]", r"\centering\scriptsize", r"\begin{threeparttable}", r"\caption{Measured multi-node benchmarks of the protocol (logistic regression, $N = 2^{22}$, medians over repeats): wall-clock of the local phase (slowest worker) and of the synchronized derivative phases, worker compute times, 99th-percentile reply latency, and bytes moved.}\label{tab:bench}",
172+ r"\begin{tabular}{@{}lrrrrrlrrrrrrrr@{}}", r"\toprule", r"Design & $K$ & hosts & $p$ & $r$ & $q$ & curvature & local (s) & fit max (s) & rounds (s) & per round (s) & der.\ max (s) & lat.\ p99 (s) & up (kB) & down (kB) \\", r"\midrule"]
173+ for r in g.itertuples():
174+ lines.append(f"{r.label.replace('_', ' ')} & {r.K} & {r.hosts} & {r.p} & {r.r} & {r.q} & {r.hessian} & {r.t_local:.3f} & {r.fit_max:.3f} & {r.t_rounds:.3f} & {r.round_wall:.3f} & {r.der_max:.3f} & {r.lat99:.3f} & {r.up_B/1e3:.1f} & {r.down_B/1e3:.1f} \\\\")
175+ lines += [r"\bottomrule", r"\end{tabular}", r"\begin{tablenotes}\footnotesize", r"\item Python 3.9/3.14 + NumPy over TCP (length-prefixed pickle frames) on the MacLustr LAN (1 GbE/10 GbE mixed); coordinator on a Mac Studio M4 Max. `local' is the wall-clock from the first `fit' broadcast to the last upload, i.e.\ the slowest worker; `rounds' sums the synchronized derivative phases. Bytes include frame headers and pickle overhead. Repeats and raw logs: \texttt{results/raw/benchmarks.jsonl}.", r"\end{tablenotes}", r"\end{threeparttable}", r"\end{table}"]
176+ (TAB / fname).write_text("\n".join(lines) + "\n"); return g
177+
178+
179+def manifest():
180+ rows = []
181+ for f in sorted(glob.glob(str(RAW / "*.manifest.json"))):
182+ m = json.load(open(f)); c = m["cell"]
183+ rows.append({"run_id": m["run_id"], "git_commit": m["git_commit"], "python_version": m["python"], "numpy": m["numpy"], "dependency_lock_sha256": "", "seed": m["seed"], "N": c["N"], "K": c["K"], "m": c["m"], "p": c["p"],
184+ "DGP": json.dumps(c["dgp"]), "method": "estimate_all", "optimizer": "batched Newton, step halving, tol 1e-9 (central 1e-12)", "tolerance": "1e-9", "replicates": m["reps"], "parallel_backend": "xargs -P / nohup", "node_count": 1,
185+ "hardware": m["host"] + " " + m["machine"], "OS": "macOS", "wall_seconds": round(m["wall_seconds"], 1), "communicated_bytes": 0, "rounds": "in-memory (no network)", "status": m["status"], "raw_output_sha256": m["raw_output_sha256"]})
186+ pd.DataFrame(rows).to_csv(ROOT / "experiments" / "EXPERIMENT_MANIFEST.csv", index=False)
187+ return len(rows)
188+
189+
190+def main():
191+ s = pd.read_csv(PROC / "summary.csv") if (PROC / "summary.csv").exists() else pd.DataFrame()
192+ made = {}
193+ if not s.empty:
194+ for (label, p), d in s.groupby(["label", "p"]):
195+ N = d.N.max()
196+ made[(label, p)] = table_cell(s, label, p, N, sorted(d.m.unique()), f"mc_{label}_p{p}.tex", f"Monte Carlo, {label.replace('_', ' ')} design, $p = {p}$, $N = {N:,}$: scaled gap to $\\widehat\\theta_N$, scaled RMSE and 95\\% coverage by local sample size $m$.")
197+ table_rates(s)
198+ table_boundary(); table_ablation("exponential", "abl_exponential.tex"); table_ablation("logistic", "abl_logistic.tex"); table_skew(); table_bench()
199+ n = manifest()
200+ # numbers macros
201+ macros = {"numMCcells": str(s.groupby(["label", "p", "N", "K"]).ngroups if not s.empty else 0), "numManifest": str(n)}
202+ if not s.empty:
203+ macros["numMCrepsExp"] = str(int(s[s.label == "exp"].reps.max())) if (s.label == "exp").any() else "0"
204+ macros["numMCrepsLogit"] = str(int(s[s.label == "logit"].reps.max())) if (s.label == "logit").any() else "0"
205+ bf = PROC / "benchmarks_summary.csv"
206+ macros["benchHosts"] = str(int(pd.read_csv(bf).hosts.max())) if bf.exists() else "0"
207+ with open(TAB / "numbers.tex", "w") as f:
208+ f.write("% Auto-generated by experiments/make_tables.py\n")
209+ for k, v in macros.items():
210+ f.write(f"\\newcommand{{\\{k}}}{{{v}}}\n")
211+ print("tables written:", sorted(p.name for p in TAB.glob("*.tex")), "manifest rows:", n)
212+
213+
214+if __name__ == "__main__":
215+ main()
modified experiments/process_results.py +8 −0
@@ -163,6 +163,9 @@ def fig_boundary(fname="fig_boundary_exponential.pdf"):
163 if not f.exists(): 163 if not f.exists():
164 return 164 return
165 b = pd.read_csv(f) 165 b = pd.read_csv(f)
166+ fp = PROC / "boundary_summary_with_prediction.csv"
167+ if fp.exists():
168+ b = pd.read_csv(fp)
166 cases = [(2, 1), (1, 2), (3, 1), (2, 0)] 169 cases = [(2, 1), (1, 2), (3, 1), (2, 0)]
167 fig, axes = plt.subplots(1, len(cases), figsize=(2.5 * len(cases) + 0.5, 2.9)) 170 fig, axes = plt.subplots(1, len(cases), figsize=(2.5 * len(cases) + 0.5, 2.9))
168 cols = ["blue", "orange", "aqua"] 171 cols = ["blue", "orange", "aqua"]
@@ -172,8 +175,13 @@ def fig_boundary(fname="fig_boundary_exponential.pdf"):
172 dd = d[d.c == c].sort_values("N") 175 dd = d[d.c == c].sort_values("N")
173 ax.errorbar(dd["N"], dd["mean_gap"], yerr=1.96 * dd["se"], marker="o", ms=3.5, color=PAL[cols[i]], label=f"c = {c:g}", capsize=2, lw=1.2) 176 ax.errorbar(dd["N"], dd["mean_gap"], yerr=1.96 * dd["se"], marker="o", ms=3.5, color=PAL[cols[i]], label=f"c = {c:g}", capsize=2, lw=1.2)
174 ax.plot(dd["N"], dd["limit_c_eff"], ls="--", color=PAL[cols[i]], lw=1.0) 177 ax.plot(dd["N"], dd["limit_c_eff"], ls="--", color=PAL[cols[i]], lw=1.0)
178+ if "pred_finite_m" in dd:
179+ ax.plot(dd["N"], dd["pred_finite_m"], ls=":", color=PAL[cols[i]], lw=1.0)
175 ax.set_xscale("log", base=2); ax.set_xlabel("N"); ax.set_title(f"(r, q) = ({r}, {q}), α = {1-1/(2**(q+1)*r):.4g}", fontsize=8, loc="left") 180 ax.set_xscale("log", base=2); ax.set_xlabel("N"); ax.set_title(f"(r, q) = ({r}, {q}), α = {1-1/(2**(q+1)*r):.4g}", fontsize=8, loc="left")
176 axes[0].set_ylabel("mean of √N (θ̃ − θ̂_N)"); axes[0].legend(fontsize=7) 181 axes[0].set_ylabel("mean of √N (θ̃ − θ̂_N)"); axes[0].legend(fontsize=7)
182+ axes[-1].text(0.02, 0.02, "solid: Monte Carlo ±1.96 s.e.; dashed: limit; dotted: finite-m prediction", transform=axes[-1].transAxes, fontsize=6, color=INK2)
183+ for ax in axes:
184+ ax.set_yscale("symlog", linthresh=1.0)
177 fig.savefig(FIG / fname); plt.close(fig) 185 fig.savefig(FIG / fname); plt.close(fig)
178 186
179 187
modified paper/main.pdf +0 −0

Binary file not shown.

added paper/tables/abl_exponential.tex +143 −0
@@ -0,0 +1,143 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Ablation (exponential, $N = 1,572,864$): RMS of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$ by split pattern, order $r$, Newton rounds $q$ and curvature used (exact / stale averaged local / single machine); quantized exact curvature with $b$ mantissa bits for $q = 1$.}\label{tab:abl_exponential}
5+\begin{tabular}{@{}llrr@{}}
6+\toprule
7+Pattern & curvature & $q$ & $m = 96$ \\
8+\midrule
9+(1) & exact & 0 & 13.223 \\
10+(1) & exact & 1 & 0.139 \\
11+(1) & exact & 2 & 1.55e-5 \\
12+(1) & exact & 3 & 2.42e-13 \\
13+(1) & stale & 1 & 0.273 \\
14+(1) & stale & 2 & 0.003 \\
15+(1) & stale & 3 & 2.98e-5 \\
16+(1) & single & 1 & 0.139 \\
17+(1) & single & 2 & 1.55e-5 \\
18+(1) & single & 3 & 2.42e-13 \\
19+(1,2) & exact & 0 & 0.323 \\
20+(1,2) & exact & 1 & 1.29e-4 \\
21+(1,2) & exact & 2 & 3.32e-11 \\
22+(1,2) & exact & 3 & 1.57e-13 \\
23+(1,2) & stale & 1 & 0.003 \\
24+(1,2) & stale & 2 & 3.27e-5 \\
25+(1,2) & stale & 3 & 3.35e-7 \\
26+(1,2) & single & 1 & 1.29e-4 \\
27+(1,2) & single & 2 & 3.32e-11 \\
28+(1,2) & single & 3 & 1.57e-13 \\
29+(1,3) & exact & 0 & 0.437 \\
30+(1,3) & exact & 1 & 1.94e-4 \\
31+(1,3) & exact & 2 & 5.28e-11 \\
32+(1,3) & exact & 3 & 1.64e-13 \\
33+(1,2,4) & exact & 0 & 0.316 \\
34+(1,2,4) & exact & 1 & 1.30e-4 \\
35+(1,2,4) & exact & 2 & 3.44e-11 \\
36+(1,2,4) & exact & 3 & 1.51e-13 \\
37+(1,2,4) & stale & 1 & 0.003 \\
38+(1,2,4) & stale & 2 & 3.43e-5 \\
39+(1,2,4) & stale & 3 & 3.55e-7 \\
40+(1,2,4) & single & 1 & 1.30e-4 \\
41+(1,2,4) & single & 2 & 3.44e-11 \\
42+(1,2,4) & single & 3 & 1.51e-13 \\
43+(1,2,3) & exact & 0 & 0.446 \\
44+(1,2,3) & exact & 1 & 2.70e-4 \\
45+(1,2,3) & exact & 2 & 1.75e-10 \\
46+(1,2,3) & exact & 3 & 1.54e-13 \\
47+(1,2,4,8) & exact & 0 & 0.387 \\
48+(1,2,4,8) & exact & 1 & 1.98e-4 \\
49+(1,2,4,8) & exact & 2 & 8.24e-11 \\
50+(1,2,4,8) & exact & 3 & 1.53e-13 \\
51+(1,2,3,4) & exact & 0 & 1.003 \\
52+(1,2,3,4) & exact & 1 & 0.001 \\
53+(1,2,3,4) & exact & 2 & 5.56e-9 \\
54+(1,2,3,4) & exact & 3 & 1.43e-13 \\
55+\midrule
56+(1) & $b$-bit Hessian, $b=2$ & 1 & 0.140 \\
57+(1) & $b$-bit Hessian, $b=3$ & 1 & 0.140 \\
58+(1) & $b$-bit Hessian, $b=4$ & 1 & 0.140 \\
59+(1) & $b$-bit Hessian, $b=6$ & 1 & 0.091 \\
60+(1) & $b$-bit Hessian, $b=8$ & 1 & 0.140 \\
61+(1) & $b$-bit Hessian, $b=12$ & 1 & 0.139 \\
62+(1) & $b$-bit Hessian, $b=16$ & 1 & 0.139 \\
63+(1) & $b$-bit Hessian, $b=24$ & 1 & 0.139 \\
64+(1) & $b$-bit Hessian, $b=52$ & 1 & 0.139 \\
65+(1) & $b$-bit score, $b=2$ & 1 & 1.792 \\
66+(1) & $b$-bit score, $b=3$ & 1 & 0.728 \\
67+(1) & $b$-bit score, $b=4$ & 1 & 0.522 \\
68+(1) & $b$-bit score, $b=6$ & 1 & 0.165 \\
69+(1) & $b$-bit score, $b=8$ & 1 & 0.142 \\
70+(1) & $b$-bit score, $b=12$ & 1 & 0.139 \\
71+(1) & $b$-bit score, $b=16$ & 1 & 0.139 \\
72+(1) & $b$-bit score, $b=24$ & 1 & 0.139 \\
73+(1) & $b$-bit score, $b=52$ & 1 & 0.139 \\
74+(1) & $b$-bit both, $b=2$ & 1 & 1.482 \\
75+(1) & $b$-bit both, $b=3$ & 1 & 0.985 \\
76+(1) & $b$-bit both, $b=4$ & 1 & 0.310 \\
77+(1) & $b$-bit both, $b=6$ & 1 & 0.129 \\
78+(1) & $b$-bit both, $b=8$ & 1 & 0.143 \\
79+(1) & $b$-bit both, $b=12$ & 1 & 0.139 \\
80+(1) & $b$-bit both, $b=16$ & 1 & 0.139 \\
81+(1) & $b$-bit both, $b=24$ & 1 & 0.139 \\
82+(1) & $b$-bit both, $b=52$ & 1 & 0.139 \\
83+(1,2) & $b$-bit Hessian, $b=2$ & 1 & 5.20e-4 \\
84+(1,2) & $b$-bit Hessian, $b=3$ & 1 & 5.20e-4 \\
85+(1,2) & $b$-bit Hessian, $b=4$ & 1 & 5.20e-4 \\
86+(1,2) & $b$-bit Hessian, $b=6$ & 1 & 5.20e-4 \\
87+(1,2) & $b$-bit Hessian, $b=8$ & 1 & 4.80e-4 \\
88+(1,2) & $b$-bit Hessian, $b=12$ & 1 & 1.29e-4 \\
89+(1,2) & $b$-bit Hessian, $b=16$ & 1 & 1.29e-4 \\
90+(1,2) & $b$-bit Hessian, $b=24$ & 1 & 1.29e-4 \\
91+(1,2) & $b$-bit Hessian, $b=52$ & 1 & 1.29e-4 \\
92+(1,2) & $b$-bit score, $b=2$ & 1 & 0.034 \\
93+(1,2) & $b$-bit score, $b=3$ & 1 & 0.017 \\
94+(1,2) & $b$-bit score, $b=4$ & 1 & 0.008 \\
95+(1,2) & $b$-bit score, $b=6$ & 1 & 0.002 \\
96+(1,2) & $b$-bit score, $b=8$ & 1 & 5.35e-4 \\
97+(1,2) & $b$-bit score, $b=12$ & 1 & 1.37e-4 \\
98+(1,2) & $b$-bit score, $b=16$ & 1 & 1.29e-4 \\
99+(1,2) & $b$-bit score, $b=24$ & 1 & 1.29e-4 \\
100+(1,2) & $b$-bit score, $b=52$ & 1 & 1.29e-4 \\
101+(1,2) & $b$-bit both, $b=2$ & 1 & 0.034 \\
102+(1,2) & $b$-bit both, $b=3$ & 1 & 0.017 \\
103+(1,2) & $b$-bit both, $b=4$ & 1 & 0.008 \\
104+(1,2) & $b$-bit both, $b=6$ & 1 & 0.002 \\
105+(1,2) & $b$-bit both, $b=8$ & 1 & 6.71e-4 \\
106+(1,2) & $b$-bit both, $b=12$ & 1 & 1.38e-4 \\
107+(1,2) & $b$-bit both, $b=16$ & 1 & 1.29e-4 \\
108+(1,2) & $b$-bit both, $b=24$ & 1 & 1.29e-4 \\
109+(1,2) & $b$-bit both, $b=52$ & 1 & 1.29e-4 \\
110+(1,2,4) & $b$-bit Hessian, $b=2$ & 1 & 4.74e-4 \\
111+(1,2,4) & $b$-bit Hessian, $b=3$ & 1 & 4.74e-4 \\
112+(1,2,4) & $b$-bit Hessian, $b=4$ & 1 & 4.74e-4 \\
113+(1,2,4) & $b$-bit Hessian, $b=6$ & 1 & 4.74e-4 \\
114+(1,2,4) & $b$-bit Hessian, $b=8$ & 1 & 4.33e-4 \\
115+(1,2,4) & $b$-bit Hessian, $b=12$ & 1 & 1.41e-4 \\
116+(1,2,4) & $b$-bit Hessian, $b=16$ & 1 & 1.30e-4 \\
117+(1,2,4) & $b$-bit Hessian, $b=24$ & 1 & 1.30e-4 \\
118+(1,2,4) & $b$-bit Hessian, $b=52$ & 1 & 1.30e-4 \\
119+(1,2,4) & $b$-bit score, $b=2$ & 1 & 0.033 \\
120+(1,2,4) & $b$-bit score, $b=3$ & 1 & 0.017 \\
121+(1,2,4) & $b$-bit score, $b=4$ & 1 & 0.008 \\
122+(1,2,4) & $b$-bit score, $b=6$ & 1 & 0.002 \\
123+(1,2,4) & $b$-bit score, $b=8$ & 1 & 5.47e-4 \\
124+(1,2,4) & $b$-bit score, $b=12$ & 1 & 1.33e-4 \\
125+(1,2,4) & $b$-bit score, $b=16$ & 1 & 1.30e-4 \\
126+(1,2,4) & $b$-bit score, $b=24$ & 1 & 1.30e-4 \\
127+(1,2,4) & $b$-bit score, $b=52$ & 1 & 1.30e-4 \\
128+(1,2,4) & $b$-bit both, $b=2$ & 1 & 0.033 \\
129+(1,2,4) & $b$-bit both, $b=3$ & 1 & 0.017 \\
130+(1,2,4) & $b$-bit both, $b=4$ & 1 & 0.008 \\
131+(1,2,4) & $b$-bit both, $b=6$ & 1 & 0.002 \\
132+(1,2,4) & $b$-bit both, $b=8$ & 1 & 7.26e-4 \\
133+(1,2,4) & $b$-bit both, $b=12$ & 1 & 1.44e-4 \\
134+(1,2,4) & $b$-bit both, $b=16$ & 1 & 1.30e-4 \\
135+(1,2,4) & $b$-bit both, $b=24$ & 1 & 1.30e-4 \\
136+(1,2,4) & $b$-bit both, $b=52$ & 1 & 1.30e-4 \\
137+\bottomrule
138+\end{tabular}
139+\begin{tablenotes}\footnotesize
140+\item 1000 paired replications per column; $K = N/m$. Rows are omitted when the pattern does not divide $m$. Source: \texttt{results/processed/ablation\_exponential\_summary.csv}.
141+\end{tablenotes}
142+\end{threeparttable}
143+\end{table}
added paper/tables/abl_logistic.tex +143 −0
@@ -0,0 +1,143 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Ablation (logistic, $N = 196,608$): RMS of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$ by split pattern, order $r$, Newton rounds $q$ and curvature used (exact / stale averaged local / single machine); quantized exact curvature with $b$ mantissa bits for $q = 1$.}\label{tab:abl_logistic}
5+\begin{tabular}{@{}llrr@{}}
6+\toprule
7+Pattern & curvature & $q$ & $m = 96$ \\
8+\midrule
9+(1) & exact & 0 & 57.950 \\
10+(1) & exact & 1 & 3.417 \\
11+(1) & exact & 2 & 0.011 \\
12+(1) & exact & 3 & 1.24e-7 \\
13+(1) & stale & 1 & 1.302 \\
14+(1) & stale & 2 & 0.109 \\
15+(1) & stale & 3 & 0.009 \\
16+(1) & single & 1 & 17.275 \\
17+(1) & single & 2 & 11.674 \\
18+(1) & single & 3 & 10.562 \\
19+(1,2) & exact & 0 & 426.941 \\
20+(1,2) & exact & 1 & 2316.233 \\
21+(1,2) & exact & 2 & 9.35e+5 \\
22+(1,2) & exact & 3 & 1.61e+14 \\
23+(1,2) & stale & 1 & 139.167 \\
24+(1,2) & stale & 2 & 33.349 \\
25+(1,2) & stale & 3 & 3.160 \\
26+(1,2) & single & 1 & 2485.332 \\
27+(1,2) & single & 2 & 2.99e+6 \\
28+(1,2) & single & 3 & 5.18e+38 \\
29+(1,3) & exact & 0 & 4391.045 \\
30+(1,3) & exact & 1 & 3.31e+5 \\
31+(1,3) & exact & 2 & 5.73e+11 \\
32+(1,3) & exact & 3 & 7.08e+109 \\
33+(1,2,4) & exact & 0 & 4.32e+4 \\
34+(1,2,4) & exact & 1 & 6.04e+9 \\
35+(1,2,4) & exact & 2 & 4.29e+16 \\
36+(1,2,4) & exact & 3 & 4.32e+16 \\
37+(1,2,4) & stale & 1 & 4.30e+4 \\
38+(1,2,4) & stale & 2 & 4.27e+4 \\
39+(1,2,4) & stale & 3 & 4.25e+4 \\
40+(1,2,4) & single & 1 & 3.48e+26 \\
41+(1,2,4) & single & 2 & 5.80e+97 \\
42+(1,2,4) & single & 3 & 5.83e+97 \\
43+(1,2,3) & exact & 0 & 8352.835 \\
44+(1,2,3) & exact & 1 & 1.02e+6 \\
45+(1,2,3) & exact & 2 & 4.99e+13 \\
46+(1,2,3) & exact & 3 & 3.80e+106 \\
47+(1,2,4,8) & exact & 0 & 8.62e+4 \\
48+(1,2,4,8) & exact & 1 & 5.25e+10 \\
49+(1,2,4,8) & exact & 2 & 1.96e+24 \\
50+(1,2,4,8) & exact & 3 & 1.96e+24 \\
51+(1,2,3,4) & exact & 0 & 1.25e+5 \\
52+(1,2,3,4) & exact & 1 & 2.92e+11 \\
53+(1,2,3,4) & exact & 2 & 3.69e+23 \\
54+(1,2,3,4) & exact & 3 & 1.83e+89 \\
55+\midrule
56+(1) & $b$-bit Hessian, $b=2$ & 1 & 3.384 \\
57+(1) & $b$-bit Hessian, $b=3$ & 1 & 3.532 \\
58+(1) & $b$-bit Hessian, $b=4$ & 1 & 3.429 \\
59+(1) & $b$-bit Hessian, $b=6$ & 1 & 3.504 \\
60+(1) & $b$-bit Hessian, $b=8$ & 1 & 3.419 \\
61+(1) & $b$-bit Hessian, $b=12$ & 1 & 3.417 \\
62+(1) & $b$-bit Hessian, $b=16$ & 1 & 3.417 \\
63+(1) & $b$-bit Hessian, $b=24$ & 1 & 3.417 \\
64+(1) & $b$-bit Hessian, $b=52$ & 1 & 3.417 \\
65+(1) & $b$-bit score, $b=2$ & 1 & 4.460 \\
66+(1) & $b$-bit score, $b=3$ & 1 & 4.101 \\
67+(1) & $b$-bit score, $b=4$ & 1 & 3.687 \\
68+(1) & $b$-bit score, $b=6$ & 1 & 3.406 \\
69+(1) & $b$-bit score, $b=8$ & 1 & 3.421 \\
70+(1) & $b$-bit score, $b=12$ & 1 & 3.418 \\
71+(1) & $b$-bit score, $b=16$ & 1 & 3.417 \\
72+(1) & $b$-bit score, $b=24$ & 1 & 3.417 \\
73+(1) & $b$-bit score, $b=52$ & 1 & 3.417 \\
74+(1) & $b$-bit both, $b=2$ & 1 & 4.755 \\
75+(1) & $b$-bit both, $b=3$ & 1 & 4.303 \\
76+(1) & $b$-bit both, $b=4$ & 1 & 3.667 \\
77+(1) & $b$-bit both, $b=6$ & 1 & 3.489 \\
78+(1) & $b$-bit both, $b=8$ & 1 & 3.423 \\
79+(1) & $b$-bit both, $b=12$ & 1 & 3.418 \\
80+(1) & $b$-bit both, $b=16$ & 1 & 3.417 \\
81+(1) & $b$-bit both, $b=24$ & 1 & 3.417 \\
82+(1) & $b$-bit both, $b=52$ & 1 & 3.417 \\
83+(1,2) & $b$-bit Hessian, $b=2$ & 1 & 2721.153 \\
84+(1,2) & $b$-bit Hessian, $b=3$ & 1 & 2622.352 \\
85+(1,2) & $b$-bit Hessian, $b=4$ & 1 & 2278.291 \\
86+(1,2) & $b$-bit Hessian, $b=6$ & 1 & 2394.384 \\
87+(1,2) & $b$-bit Hessian, $b=8$ & 1 & 2305.622 \\
88+(1,2) & $b$-bit Hessian, $b=12$ & 1 & 2317.290 \\
89+(1,2) & $b$-bit Hessian, $b=16$ & 1 & 2316.103 \\
90+(1,2) & $b$-bit Hessian, $b=24$ & 1 & 2316.233 \\
91+(1,2) & $b$-bit Hessian, $b=52$ & 1 & 2316.233 \\
92+(1,2) & $b$-bit score, $b=2$ & 1 & 2443.159 \\
93+(1,2) & $b$-bit score, $b=3$ & 1 & 2241.467 \\
94+(1,2) & $b$-bit score, $b=4$ & 1 & 2359.819 \\
95+(1,2) & $b$-bit score, $b=6$ & 1 & 2305.944 \\
96+(1,2) & $b$-bit score, $b=8$ & 1 & 2316.935 \\
97+(1,2) & $b$-bit score, $b=12$ & 1 & 2316.057 \\
98+(1,2) & $b$-bit score, $b=16$ & 1 & 2316.229 \\
99+(1,2) & $b$-bit score, $b=24$ & 1 & 2316.233 \\
100+(1,2) & $b$-bit score, $b=52$ & 1 & 2316.233 \\
101+(1,2) & $b$-bit both, $b=2$ & 1 & 2852.513 \\
102+(1,2) & $b$-bit both, $b=3$ & 1 & 2537.858 \\
103+(1,2) & $b$-bit both, $b=4$ & 1 & 2321.353 \\
104+(1,2) & $b$-bit both, $b=6$ & 1 & 2383.708 \\
105+(1,2) & $b$-bit both, $b=8$ & 1 & 2306.320 \\
106+(1,2) & $b$-bit both, $b=12$ & 1 & 2317.114 \\
107+(1,2) & $b$-bit both, $b=16$ & 1 & 2316.099 \\
108+(1,2) & $b$-bit both, $b=24$ & 1 & 2316.233 \\
109+(1,2) & $b$-bit both, $b=52$ & 1 & 2316.233 \\
110+(1,2,4) & $b$-bit Hessian, $b=2$ & 1 & 1.22e+7 \\
111+(1,2,4) & $b$-bit Hessian, $b=3$ & 1 & 4.70e+6 \\
112+(1,2,4) & $b$-bit Hessian, $b=4$ & 1 & 6.75e+7 \\
113+(1,2,4) & $b$-bit Hessian, $b=6$ & 1 & 1.63e+7 \\
114+(1,2,4) & $b$-bit Hessian, $b=8$ & 1 & 3.20e+7 \\
115+(1,2,4) & $b$-bit Hessian, $b=12$ & 1 & 9.10e+8 \\
116+(1,2,4) & $b$-bit Hessian, $b=16$ & 1 & 7.52e+9 \\
117+(1,2,4) & $b$-bit Hessian, $b=24$ & 1 & 6.03e+9 \\
118+(1,2,4) & $b$-bit Hessian, $b=52$ & 1 & 6.04e+9 \\
119+(1,2,4) & $b$-bit score, $b=2$ & 1 & 5.76e+9 \\
120+(1,2,4) & $b$-bit score, $b=3$ & 1 & 5.92e+9 \\
121+(1,2,4) & $b$-bit score, $b=4$ & 1 & 5.92e+9 \\
122+(1,2,4) & $b$-bit score, $b=6$ & 1 & 6.06e+9 \\
123+(1,2,4) & $b$-bit score, $b=8$ & 1 & 6.03e+9 \\
124+(1,2,4) & $b$-bit score, $b=12$ & 1 & 6.04e+9 \\
125+(1,2,4) & $b$-bit score, $b=16$ & 1 & 6.04e+9 \\
126+(1,2,4) & $b$-bit score, $b=24$ & 1 & 6.04e+9 \\
127+(1,2,4) & $b$-bit score, $b=52$ & 1 & 6.04e+9 \\
128+(1,2,4) & $b$-bit both, $b=2$ & 1 & 1.20e+7 \\
129+(1,2,4) & $b$-bit both, $b=3$ & 1 & 4.69e+6 \\
130+(1,2,4) & $b$-bit both, $b=4$ & 1 & 6.79e+7 \\
131+(1,2,4) & $b$-bit both, $b=6$ & 1 & 1.63e+7 \\
132+(1,2,4) & $b$-bit both, $b=8$ & 1 & 3.20e+7 \\
133+(1,2,4) & $b$-bit both, $b=12$ & 1 & 9.10e+8 \\
134+(1,2,4) & $b$-bit both, $b=16$ & 1 & 7.52e+9 \\
135+(1,2,4) & $b$-bit both, $b=24$ & 1 & 6.03e+9 \\
136+(1,2,4) & $b$-bit both, $b=52$ & 1 & 6.04e+9 \\
137+\bottomrule
138+\end{tabular}
139+\begin{tablenotes}\footnotesize
140+\item 80 paired replications per column; $K = N/m$. Rows are omitted when the pattern does not divide $m$. Source: \texttt{results/processed/ablation\_logistic\_summary.csv}.
141+\end{tablenotes}
142+\end{threeparttable}
143+\end{table}
added paper/tables/bench.tex +1 −0
@@ -0,0 +1 @@
1+% no benchmark data
added paper/tables/boundary.tex +90 −0
@@ -0,0 +1,90 @@
1+\begin{table}[t]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Boundary study in the exponential model (Theorem~\ref{thm:exponential}(v)): mean of $\sqrt N(\theta^{(q)}_r - \widehat\theta_N)$ when $K \approx cN^{\alpha(r,q)}$, against the limit $-\theta_0(\beta_r c_{\mathrm{eff}}^r)^{2^q}$ ($q\ge1$) or $\beta_r\theta_0 c_{\mathrm{eff}}^r$ ($q=0$), with $c_{\mathrm{eff}} = K/N^{\alpha}$ after rounding $m$ to a multiple of the finest split.}\label{tab:boundary}
5+\begin{tabular}{@{}ccrrrrrrrr@{}}
6+\toprule
7+$(r,q)$ & $c$ & $N$ & $K$ & $m$ & reps & mean gap (s.e.) & finite-$m$ prediction & limit at $c_{\mathrm{eff}}$ & gap/limit \\
8+\midrule
9+(1,1) & 0.5 & $2^{16}$ & 2,048 & 32 & 2000 & -0.266 (0.00039) & -0.2667 & -0.25 & 1.06 \\
10+(1,1) & 0.5 & $2^{18}$ & 4,096 & 64 & 2000 & -0.1289 (0.00013) & -0.1291 & -0.125 & 1.03 \\
11+(1,1) & 0.5 & $2^{20}$ & 16,384 & 64 & 2000 & -0.2582 (0.00013) & -0.258 & -0.25 & 1.03 \\
12+(1,1) & 0.5 & $2^{22}$ & 32,768 & 128 & 1000 & -0.127 (6.4e-05) & -0.127 & -0.125 & 1.02 \\
13+(1,1) & 0.5 & $2^{24}$ & 131,072 & 128 & 400 & -0.2541 (0.0001) & -0.254 & -0.25 & 1.02 \\
14+(1,1) & 1 & $2^{16}$ & 4,096 & 16 & 2000 & -1.138 (0.0012) & -1.138 & -1 & 1.14 \\
15+(1,1) & 1 & $2^{18}$ & 8,192 & 32 & 2000 & -0.5328 (0.00039) & -0.5329 & -0.5 & 1.07 \\
16+(1,1) & 1 & $2^{20}$ & 32,768 & 32 & 2000 & -1.065 (0.00038) & -1.066 & -1 & 1.07 \\
17+(1,1) & 1 & $2^{22}$ & 65,536 & 64 & 1000 & -0.5158 (0.00019) & -0.516 & -0.5 & 1.03 \\
18+(1,1) & 1 & $2^{24}$ & 262,144 & 64 & 400 & -1.032 (0.00029) & -1.032 & -1 & 1.03 \\
19+(1,1) & 2 & $2^{16}$ & 8,192 & 8 & 2000 & -5.227 (0.0043) & -5.226 & -4 & 1.31 \\
20+(1,1) & 2 & $2^{18}$ & 16,384 & 16 & 2000 & -2.274 (0.0012) & -2.276 & -2 & 1.14 \\
21+(1,1) & 2 & $2^{20}$ & 65,536 & 16 & 2000 & -4.55 (0.0012) & -4.551 & -4 & 1.14 \\
22+(1,1) & 2 & $2^{22}$ & 131,072 & 32 & 1000 & -2.13 (0.00052) & -2.131 & -2 & 1.07 \\
23+(1,1) & 2 & $2^{24}$ & 524,288 & 32 & 400 & -4.262 (0.00088) & -4.262 & -4 & 1.07 \\
24+(1,2) & 0.5 & $2^{16}$ & 8,192 & 8 & 2000 & -0.1068 (0.00018) & -0.1068 & -0.0625 & 1.71 \\
25+(1,2) & 0.5 & $2^{18}$ & 16,384 & 16 & 2000 & -0.01011 (1.1e-05) & -0.01012 & -0.007812 & 1.29 \\
26+(1,2) & 0.5 & $2^{20}$ & 65,536 & 16 & 2000 & -0.02023 (1.1e-05) & -0.02023 & -0.01562 & 1.29 \\
27+(1,2) & 0.5 & $2^{22}$ & 262,144 & 16 & 1000 & -0.04046 (1.6e-05) & -0.04046 & -0.03125 & 1.29 \\
28+(1,2) & 0.5 & $2^{24}$ & 1,048,576 & 16 & 400 & -0.08092 (2.3e-05) & -0.08091 & -0.0625 & 1.29 \\
29+(1,2) & 1 & $2^{16}$ & 16,384 & 4 & 2000 & -3.163 (0.0045) & -3.165 & -1 & 3.16 \\
30+(1,2) & 1 & $2^{18}$ & 32,768 & 8 & 2000 & -0.2131 (0.00017) & -0.2134 & -0.125 & 1.71 \\
31+(1,2) & 1 & $2^{20}$ & 131,072 & 8 & 2000 & -0.4264 (0.00017) & -0.4265 & -0.25 & 1.71 \\
32+(1,2) & 1 & $2^{22}$ & 524,288 & 8 & 1000 & -0.8533 (0.00025) & -0.853 & -0.5 & 1.71 \\
33+(1,2) & 1 & $2^{24}$ & 2,097,152 & 8 & 400 & -1.705 (0.00038) & -1.706 & -1 & 1.71 \\
34+(1,2) & 2 & $2^{16}$ & 32,768 & 2 & 2000 & -257.5 (0.92) & -- & -16 & 16.09 \\
35+(1,2) & 2 & $2^{18}$ & 131,072 & 2 & 2000 & -513.1 (0.74) & -- & -32 & 16.03 \\
36+(1,2) & 2 & $2^{20}$ & 262,144 & 4 & 2000 & -12.64 (0.0044) & -12.64 & -4 & 3.16 \\
37+(1,2) & 2 & $2^{22}$ & 1,048,576 & 4 & 1000 & -25.29 (0.006) & -25.28 & -8 & 3.16 \\
38+(1,2) & 2 & $2^{24}$ & 4,194,304 & 4 & 400 & -50.55 (0.01) & -50.57 & -16 & 3.16 \\
39+(2,0) & 0.5 & $2^{16}$ & 2,048 & 32 & 2000 & -0.5522 (0.0085) & -0.5505 & -0.5 & 1.10 \\
40+(2,0) & 0.5 & $2^{18}$ & 4,096 & 64 & 2000 & -0.2589 (0.0058) & -0.2622 & -0.25 & 1.04 \\
41+(2,0) & 0.5 & $2^{20}$ & 16,384 & 64 & 2000 & -0.5192 (0.0058) & -0.5243 & -0.5 & 1.04 \\
42+(2,0) & 0.5 & $2^{22}$ & 32,768 & 128 & 1000 & -0.2469 (0.0056) & -0.256 & -0.25 & 0.99 \\
43+(2,0) & 0.5 & $2^{24}$ & 131,072 & 128 & 400 & -0.5122 (0.0089) & -0.5119 & -0.5 & 1.02 \\
44+(2,0) & 1 & $2^{16}$ & 4,096 & 16 & 2000 & -2.438 (0.014) & -2.438 & -2 & 1.22 \\
45+(2,0) & 1 & $2^{18}$ & 8,192 & 32 & 2000 & -1.102 (0.0086) & -1.101 & -1 & 1.10 \\
46+(2,0) & 1 & $2^{20}$ & 32,768 & 32 & 2000 & -2.208 (0.0088) & -2.202 & -2 & 1.10 \\
47+(2,0) & 1 & $2^{22}$ & 65,536 & 64 & 1000 & -1.036 (0.0085) & -1.049 & -1 & 1.04 \\
48+(2,0) & 1 & $2^{24}$ & 262,144 & 64 & 400 & -2.104 (0.014) & -2.097 & -2 & 1.05 \\
49+(2,0) & 2 & $2^{16}$ & 8,192 & 8 & 2000 & -12.19 (0.028) & -12.19 & -8 & 1.52 \\
50+(2,0) & 2 & $2^{18}$ & 16,384 & 16 & 2000 & -4.863 (0.014) & -4.876 & -4 & 1.22 \\
51+(2,0) & 2 & $2^{20}$ & 65,536 & 16 & 2000 & -9.737 (0.014) & -9.752 & -8 & 1.22 \\
52+(2,0) & 2 & $2^{22}$ & 131,072 & 32 & 1000 & -4.391 (0.013) & -4.404 & -4 & 1.10 \\
53+(2,0) & 2 & $2^{24}$ & 524,288 & 32 & 400 & -8.8 (0.018) & -8.809 & -8 & 1.10 \\
54+(2,1) & 0.5 & $2^{16}$ & 8,192 & 8 & 2000 & -0.5876 (0.0026) & -0.5863 & -0.25 & 2.35 \\
55+(2,1) & 0.5 & $2^{18}$ & 16,384 & 16 & 2000 & -0.04726 (0.00026) & -0.0472 & -0.03125 & 1.51 \\
56+(2,1) & 0.5 & $2^{20}$ & 65,536 & 16 & 2000 & -0.09306 (0.00027) & -0.09326 & -0.0625 & 1.49 \\
57+(2,1) & 0.5 & $2^{22}$ & 262,144 & 16 & 1000 & -0.1866 (0.00038) & -0.1859 & -0.125 & 1.49 \\
58+(2,1) & 0.5 & $2^{24}$ & 1,048,576 & 16 & 400 & -0.3725 (0.0006) & -0.3716 & -0.25 & 1.49 \\
59+(2,1) & 1 & $2^{16}$ & 16,384 & 4 & 2000 & -28.8 (0.17) & -- & -4 & 7.20 \\
60+(2,1) & 1 & $2^{18}$ & 65,536 & 4 & 2000 & -57.3 (0.2) & -- & -8 & 7.16 \\
61+(2,1) & 1 & $2^{20}$ & 131,072 & 8 & 2000 & -2.323 (0.0027) & -2.323 & -1 & 2.32 \\
62+(2,1) & 1 & $2^{22}$ & 524,288 & 8 & 1000 & -4.637 (0.0036) & -4.645 & -2 & 2.32 \\
63+(2,1) & 1 & $2^{24}$ & 2,097,152 & 8 & 400 & -9.287 (0.0057) & -9.288 & -4 & 2.32 \\
64+(2,1) & 2 & $2^{16}$ & 16,384 & 4 & 2000 & -28.72 (0.14) & -- & -4 & 7.18 \\
65+(2,1) & 2 & $2^{18}$ & 65,536 & 4 & 2000 & -57.07 (0.12) & -- & -8 & 7.13 \\
66+(2,1) & 2 & $2^{20}$ & 262,144 & 4 & 2000 & -113.8 (0.12) & -- & -16 & 7.11 \\
67+(2,1) & 2 & $2^{22}$ & 1,048,576 & 4 & 1000 & -227.7 (0.18) & -- & -32 & 7.12 \\
68+(2,1) & 2 & $2^{24}$ & 4,194,304 & 4 & 400 & -455.1 (0.33) & -- & -64 & 7.11 \\
69+(3,1) & 0.5 & $2^{16}$ & 8,192 & 8 & 2000 & -0.7203 (0.1) & -- & -0.0625 & 11.53 \\
70+(3,1) & 0.5 & $2^{18}$ & 32,768 & 8 & 2000 & -1.178 (0.0066) & -- & -0.125 & 9.42 \\
71+(3,1) & 0.5 & $2^{20}$ & 131,072 & 8 & 2000 & -2.339 (0.0077) & -- & -0.25 & 9.36 \\
72+(3,1) & 0.5 & $2^{22}$ & 524,288 & 8 & 1000 & -4.668 (0.012) & -- & -0.5 & 9.34 \\
73+(3,1) & 0.5 & $2^{24}$ & 2,097,152 & 8 & 400 & -9.308 (0.016) & -- & -1 & 9.31 \\
74+(3,1) & 1 & $2^{16}$ & 8,192 & 8 & 2000 & -0.6026 (0.0082) & -- & -0.0625 & 9.64 \\
75+(3,1) & 1 & $2^{18}$ & 32,768 & 8 & 2000 & -1.178 (0.0065) & -- & -0.125 & 9.42 \\
76+(3,1) & 1 & $2^{20}$ & 131,072 & 8 & 2000 & -2.328 (0.0084) & -- & -0.25 & 9.31 \\
77+(3,1) & 1 & $2^{22}$ & 524,288 & 8 & 1000 & -4.641 (0.01) & -- & -0.5 & 9.28 \\
78+(3,1) & 1 & $2^{24}$ & 2,097,152 & 8 & 400 & -9.299 (0.015) & -- & -1 & 9.30 \\
79+(3,1) & 2 & $2^{16}$ & 8,192 & 8 & 2000 & -0.6075 (0.0075) & -- & -0.0625 & 9.72 \\
80+(3,1) & 2 & $2^{18}$ & 32,768 & 8 & 2000 & -1.19 (0.011) & -- & -0.125 & 9.52 \\
81+(3,1) & 2 & $2^{20}$ & 131,072 & 8 & 2000 & -2.332 (0.0067) & -- & -0.25 & 9.33 \\
82+(3,1) & 2 & $2^{22}$ & 524,288 & 8 & 1000 & -4.65 (0.01) & -- & -0.5 & 9.30 \\
83+(3,1) & 2 & $2^{24}$ & 2,097,152 & 8 & 400 & -9.289 (0.016) & -- & -1 & 9.29 \\
84+\bottomrule
85+\end{tabular}
86+\begin{tablenotes}\footnotesize
87+\item Exact-Gamma simulation of block sufficient statistics; seed 20261028 plus a cell hash. The ratio converges to one only as $m\to\infty$; with $m \asymp N^{1/(2^{q+1}r)}$ the local sample sizes in the table are small and the finite-$m$ bias $\theta_0\sum_j w_js_j/(m-s_j)$ exceeds its leading term. Source: \texttt{results/processed/boundary\_summary.csv}.
88+\end{tablenotes}
89+\end{threeparttable}
90+\end{table}
added paper/tables/mc_exp_contam_p1.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, exp contam design, $p = 1$, $N = 1,048,576$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_exp_contam_p1}
5+\begin{tabular}{@{}lrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 16$, $K = 65,536$} \\
8+ & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 101.734 & 17.256 & 0.000 \\
11+(1,1) average + 1 Newton & 11.018 & 95.547 & 0.000 \\
12+(1,2) average + 2 Newton & 0.129 & 84.659 & 0.000 \\
13+(2,0) half-block jackknife & 12.900 & 71.658 & 0.000 \\
14+(2,1) half-block + 1 Newton & 0.179 & 84.707 & 0.000 \\
15+(2,2) half-block + 2 Newton & 3.51e-5 & 84.530 & -- \\
16+(3,0) quarter-block & 21.466 & 63.100 & -- \\
17+(3,1) quarter-block + 1 Newton & 0.493 & 85.020 & 0.000 \\
18+analytic Cox--Snell + 1 Newton & 1.432 & 85.961 & 0.000 \\
19+CSL, 1 round & 1.39e-13 & 84.530 & -- \\
20+CSL, 2 rounds & 1.35e-13 & 84.530 & -- \\
21+AEE (Lin--Xi) & 120.265 & 204.789 & -- \\
22+(2,1) stale Hessian & 1.806 & 82.728 & -- \\
23+centralized MLE & 0 & 84.530 & 0.000 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 1000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.007$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=16$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_exp_p1.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, exp design, $p = 1$, $N = 262,144$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_exp_p1}
5+\begin{tabular}{@{}lrrrrrrrrrrrrrrrrrrrrrrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 8$, $K = 32,768$} & \multicolumn{3}{c}{$m = 16$, $K = 16,384$} & \multicolumn{3}{c}{$m = 32$, $K = 8,192$} & \multicolumn{3}{c}{$m = 64$, $K = 4,096$} & \multicolumn{3}{c}{$m = 128$, $K = 2,048$} & \multicolumn{3}{c}{$m = 256$, $K = 1,024$} & \multicolumn{3}{c}{$m = 512$, $K = 512$} & \multicolumn{3}{c}{$m = 1024$, $K = 256$} \\
8+ & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 73.146 & 73.177 & 0.000 & 34.124 & 34.152 & 0.000 & 16.520 & 16.548 & 0.000 & 8.125 & 8.231 & 0.000 & 4.034 & 4.191 & 0.016 & 2.005 & 2.220 & 0.493 & 1.005 & 1.413 & 0.829 & 0.500 & 1.098 & 0.921 \\
11+(1,1) average + 1 Newton & 10.451 & 10.474 & 0.000 & 2.275 & 2.476 & 0.371 & 0.533 & 1.143 & 0.906 & 0.129 & 1.002 & 0.950 & 0.032 & 0.978 & 0.958 & 0.008 & 0.995 & 0.954 & 0.002 & 0.996 & 0.949 & 4.96e-4 & 0.994 & 0.951 \\
12+(1,2) average + 2 Newton & 0.213 & 1.008 & 0.952 & 0.010 & 1.007 & 0.948 & 5.57e-4 & 1.009 & 0.952 & 3.27e-5 & 0.999 & 0.947 & 2.00e-6 & 0.979 & 0.958 & 1.23e-7 & 0.994 & 0.956 & 7.96e-9 & 0.996 & 0.949 & 5.11e-10 & 0.994 & 0.950 \\
13+(2,0) half-block jackknife & 24.405 & 24.402 & 0.000 & 4.921 & 5.013 & 0.006 & 1.159 & 1.540 & 0.777 & 0.380 & 1.055 & 0.934 & 0.195 & 0.997 & 0.949 & 0.129 & 1.006 & 0.954 & 0.090 & 0.996 & 0.947 & 0.063 & 0.998 & 0.949 \\
14+(2,1) half-block + 1 Newton & 1.169 & 1.514 & 0.791 & 0.049 & 1.008 & 0.946 & 0.003 & 1.009 & 0.953 & 4.50e-4 & 0.999 & 0.947 & 1.25e-4 & 0.979 & 0.958 & 5.65e-5 & 0.994 & 0.956 & 2.85e-5 & 0.996 & 0.949 & 1.34e-5 & 0.994 & 0.950 \\
15+(2,2) half-block + 2 Newton & 0.003 & 0.990 & -- & 5.25e-6 & 1.007 & -- & 3.19e-8 & 1.009 & -- & 1.12e-9 & 0.999 & -- & 1.09e-10 & 0.979 & -- & 2.30e-11 & 0.994 & -- & 7.79e-12 & 0.996 & -- & 1.14e-12 & 0.994 & -- \\
16+(3,0) quarter-block & 24.625 & 24.669 & -- & 1.893 & 2.159 & -- & 0.590 & 1.169 & -- & 0.387 & 1.067 & -- & 0.265 & 1.018 & -- & 0.179 & 1.013 & -- & 0.127 & 0.998 & -- & 0.088 & 1.000 & -- \\
17+(3,1) quarter-block + 1 Newton & 1.281 & 1.602 & 0.788 & 0.010 & 1.007 & 0.948 & 0.001 & 1.009 & 0.953 & 5.03e-4 & 0.999 & 0.947 & 2.37e-4 & 0.979 & 0.958 & 1.09e-4 & 0.994 & 0.956 & 5.74e-5 & 0.996 & 0.949 & 2.62e-5 & 0.994 & 0.950 \\
18+analytic Cox--Snell + 1 Newton & 0.001 & 0.990 & 0.954 & 4.87e-4 & 1.007 & 0.948 & 2.25e-4 & 1.009 & 0.952 & 1.12e-4 & 0.999 & 0.947 & 5.57e-5 & 0.979 & 0.958 & 2.81e-5 & 0.994 & 0.956 & 1.39e-5 & 0.996 & 0.949 & 6.64e-6 & 0.994 & 0.950 \\
19+CSL, 1 round & 7.57e-14 & 0.990 & -- & 7.61e-14 & 1.007 & -- & 7.25e-14 & 1.009 & -- & 7.88e-14 & 0.999 & -- & 7.24e-14 & 0.979 & -- & 7.73e-14 & 0.994 & -- & 7.51e-14 & 0.996 & -- & 7.73e-14 & 0.994 & -- \\
20+CSL, 2 rounds & 7.88e-14 & 0.990 & -- & 7.20e-14 & 1.007 & -- & 7.59e-14 & 1.009 & -- & 7.71e-14 & 0.999 & -- & 7.24e-14 & 0.979 & -- & 7.58e-14 & 0.994 & -- & 7.63e-14 & 0.996 & -- & 8.10e-14 & 0.994 & -- \\
21+AEE (Lin--Xi) & 56.889 & 56.874 & -- & 30.110 & 30.114 & -- & 15.518 & 15.553 & -- & 7.875 & 7.892 & -- & 3.972 & 4.051 & -- & 1.990 & 2.242 & -- & 1.001 & 1.413 & -- & 0.499 & 1.127 & -- \\
22+(2,1) stale Hessian & 1.622 & 1.880 & -- & 0.243 & 1.033 & -- & 0.032 & 1.010 & -- & 0.005 & 0.999 & -- & 0.001 & 0.979 & -- & 4.86e-4 & 0.995 & -- & 1.74e-4 & 0.996 & -- & 6.30e-5 & 0.994 & -- \\
23+centralized MLE & 0 & 0.990 & 0.954 & 0 & 1.007 & 0.948 & 0 & 1.009 & 0.953 & 0 & 0.999 & 0.947 & 0 & 0.979 & 0.958 & 0 & 0.994 & 0.956 & 0 & 0.996 & 0.949 & 0 & 0.994 & 0.950 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 2000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.005$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=256$: 1.000 / 1.000; $m=16$: 1.000 / 1.000; $m=128$: 1.000 / 1.000; $m=1024$: 1.000 / 1.000; $m=8$: 1.000 / 1.000; $m=64$: 1.000 / 1.000; $m=512$: 1.000 / 1.000; $m=32$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_logit_misspec_p3.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, logit misspec design, $p = 3$, $N = 262,144$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_logit_misspec_p3}
5+\begin{tabular}{@{}lrrrrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 128$, $K = 2,048$} & \multicolumn{3}{c}{$m = 256$, $K = 1,024$} \\
8+ & gap & RMSE & cov & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 12.365 & 73.446 & 0.129 & 6.032 & 77.571 & 0.003 \\
11+(1,1) average + 1 Newton & 0.099 & 81.930 & 0.000 & 0.023 & 81.810 & 0.000 \\
12+(1,2) average + 2 Newton & 6.54e-6 & 81.854 & 0.000 & 3.65e-7 & 81.792 & 0.000 \\
13+(2,0) half-block jackknife & 1.668 & 82.831 & 0.000 & 0.714 & 82.009 & 0.000 \\
14+(2,1) half-block + 1 Newton & 0.002 & 81.856 & 0.000 & 3.61e-4 & 81.792 & 0.000 \\
15+(2,2) half-block + 2 Newton & 4.97e-9 & 81.854 & -- & 2.09e-10 & 81.792 & -- \\
16+(3,0) quarter-block & 13.789 & 81.339 & -- & 0.947 & 81.752 & -- \\
17+(3,1) quarter-block + 1 Newton & 2.251 & 81.978 & 0.000 & 6.00e-4 & 81.792 & 0.000 \\
18+analytic Cox--Snell + 1 Newton & 0.003 & 81.856 & 0.000 & 7.22e-4 & 81.792 & 0.000 \\
19+CSL, 1 round & 1.986 & 82.096 & -- & 0.666 & 81.850 & -- \\
20+CSL, 2 rounds & 0.564 & 81.610 & -- & 0.128 & 81.737 & -- \\
21+AEE (Lin--Xi) & 10.684 & 90.447 & -- & 5.413 & 86.086 & -- \\
22+(2,1) stale Hessian & 0.051 & 81.826 & -- & 0.011 & 81.789 & -- \\
23+centralized MLE & 0 & 81.854 & 0.000 & 0 & 81.792 & 0.000 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 1000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.007$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=256$: 1.000 / 1.000; $m=128$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_logit_p3.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, logit design, $p = 3$, $N = 262,144$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_logit_p3}
5+\begin{tabular}{@{}lrrrrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 128$, $K = 2,048$} & \multicolumn{3}{c}{$m = 256$, $K = 1,024$} \\
8+ & gap & RMSE & cov & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 14.498 & 14.992 & 0.175 & 7.074 & 8.186 & 0.560 \\
11+(1,1) average + 1 Newton & 0.159 & 3.821 & 0.948 & 0.038 & 3.882 & 0.947 \\
12+(1,2) average + 2 Newton & 1.89e-5 & 3.819 & 0.948 & 1.07e-6 & 3.883 & 0.946 \\
13+(2,0) half-block jackknife & 2.142 & 4.359 & 0.908 & 0.814 & 3.967 & 0.939 \\
14+(2,1) half-block + 1 Newton & 0.004 & 3.819 & 0.948 & 5.65e-4 & 3.883 & 0.946 \\
15+(2,2) half-block + 2 Newton & 1.81e-8 & 3.819 & -- & 6.37e-10 & 3.883 & -- \\
16+(3,0) quarter-block & 28.168 & 28.417 & -- & 1.033 & 4.049 & -- \\
17+(3,1) quarter-block + 1 Newton & 10.997 & 11.618 & 0.938 & 8.19e-4 & 3.883 & 0.946 \\
18+analytic Cox--Snell + 1 Newton & 3.44e-4 & 3.819 & 0.948 & 1.93e-4 & 3.883 & 0.946 \\
19+CSL, 1 round & 2.120 & 4.408 & -- & 0.710 & 3.937 & -- \\
20+CSL, 2 rounds & 0.596 & 3.849 & -- & 0.127 & 3.888 & -- \\
21+AEE (Lin--Xi) & 13.417 & 13.951 & -- & 6.794 & 7.708 & -- \\
22+(2,1) stale Hessian & 0.060 & 3.820 & -- & 0.012 & 3.883 & -- \\
23+centralized MLE & 0 & 3.819 & 0.948 & 0 & 3.883 & 0.946 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 1000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.007$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=256$: 1.000 / 1.000; $m=128$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_logit_p8.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, logit design, $p = 8$, $N = 16,384$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_logit_p8}
5+\begin{tabular}{@{}lrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 256$, $K = 64$} \\
8+ & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 5.571 & 8.708 & 0.851 \\
11+(1,1) average + 1 Newton & 0.107 & 6.382 & 0.951 \\
12+(1,2) average + 2 Newton & 4.24e-5 & 6.388 & 0.952 \\
13+(2,0) half-block jackknife & 2.187 & 6.702 & 0.938 \\
14+(2,1) half-block + 1 Newton & 0.014 & 6.388 & 0.952 \\
15+(2,2) half-block + 2 Newton & 1.21e-6 & 6.388 & -- \\
16+(3,0) quarter-block & 2.992 & 7.046 & -- \\
17+(3,1) quarter-block + 1 Newton & 0.023 & 6.387 & 0.952 \\
18+analytic Cox--Snell + 1 Newton & 0.004 & 6.388 & 0.952 \\
19+CSL, 1 round & 0.802 & 6.420 & -- \\
20+CSL, 2 rounds & 0.253 & 6.405 & -- \\
21+AEE (Lin--Xi) & 5.091 & 7.946 & -- \\
22+(2,1) stale Hessian & 0.085 & 6.391 & -- \\
23+centralized MLE & 0 & 6.388 & 0.952 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 1000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.007$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=256$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_pois_p3.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, pois design, $p = 3$, $N = 65,536$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_pois_p3}
5+\begin{tabular}{@{}lrrrrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 32$, $K = 2,048$} & \multicolumn{3}{c}{$m = 64$, $K = 1,024$} \\
8+ & gap & RMSE & cov & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 10.480 & 10.623 & 0.581 & 5.054 & 5.276 & 0.629 \\
11+(1,1) average + 1 Newton & 0.222 & 1.592 & 0.949 & 0.051 & 1.595 & 0.953 \\
12+(1,2) average + 2 Newton & 1.03e-4 & 1.580 & 0.952 & 5.45e-6 & 1.593 & 0.954 \\
13+(2,0) half-block jackknife & 2.702 & 3.111 & 0.644 & 0.879 & 1.843 & 0.904 \\
14+(2,1) half-block + 1 Newton & 0.031 & 1.580 & 0.951 & 0.002 & 1.594 & 0.954 \\
15+(2,2) half-block + 2 Newton & 5.30e-5 & 1.580 & -- & 2.05e-8 & 1.593 & -- \\
16+(3,0) quarter-block & 1199.057 & 1199.047 & -- & 1.045 & 1.931 & -- \\
17+(3,1) quarter-block + 1 Newton & 1198.464 & 1198.453 & 0.348 & 0.007 & 1.594 & 0.954 \\
18+analytic Cox--Snell + 1 Newton & 0.001 & 1.580 & 0.952 & 6.26e-4 & 1.593 & 0.954 \\
19+CSL, 1 round & 3.362 & 3.712 & -- & 1.044 & 1.907 & -- \\
20+CSL, 2 rounds & 4.386 & 4.674 & -- & 0.645 & 1.735 & -- \\
21+AEE (Lin--Xi) & 10.100 & 10.198 & -- & 4.964 & 5.235 & -- \\
22+(2,1) stale Hessian & 0.036 & 1.580 & -- & 0.008 & 1.594 & -- \\
23+centralized MLE & 0 & 1.580 & 0.952 & 0 & 1.593 & 0.954 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 1000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.007$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=64$: 1.000 / 1.000; $m=32$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_pois_p8.tex +31 −0
@@ -0,0 +1,31 @@
1+\begin{table}[p]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Monte Carlo, pois design, $p = 8$, $N = 262,144$: scaled gap to $\widehat\theta_N$, scaled RMSE and 95\% coverage by local sample size $m$.}\label{tab:mc_pois_p8}
5+\begin{tabular}{@{}lrrrrrrrrr@{}}
6+\toprule
7+Estimator & \multicolumn{3}{c}{$m = 256$, $K = 1,024$} & \multicolumn{3}{c}{$m = 512$, $K = 512$} & \multicolumn{3}{c}{$m = 1024$, $K = 256$} \\
8+ & gap & RMSE & cov & gap & RMSE & cov & gap & RMSE & cov \\
9+\midrule
10+(1,0) plain average & 6.250 & 6.728 & 0.831 & 3.103 & 4.012 & 0.841 & 1.553 & 2.938 & 0.912 \\
11+(1,1) average + 1 Newton & 0.039 & 2.516 & 0.951 & 0.010 & 2.545 & 0.946 & 0.002 & 2.511 & 0.949 \\
12+(1,2) average + 2 Newton & 1.57e-6 & 2.516 & 0.951 & 1.02e-7 & 2.545 & 0.947 & 7.28e-9 & 2.511 & 0.949 \\
13+(2,0) half-block jackknife & 0.924 & 2.672 & 0.935 & 0.575 & 2.606 & 0.939 & 0.399 & 2.543 & 0.944 \\
14+(2,1) half-block + 1 Newton & 9.83e-4 & 2.516 & 0.951 & 4.08e-4 & 2.545 & 0.947 & 1.99e-4 & 2.511 & 0.949 \\
15+(2,2) half-block + 2 Newton & 1.68e-9 & 2.516 & -- & 3.48e-10 & 2.545 & -- & 8.56e-11 & 2.511 & -- \\
16+(3,0) quarter-block & 1.183 & 2.771 & -- & 0.806 & 2.664 & -- & 0.562 & 2.578 & -- \\
17+(3,1) quarter-block + 1 Newton & 0.002 & 2.516 & 0.951 & 8.08e-4 & 2.545 & 0.947 & 3.98e-4 & 2.511 & 0.949 \\
18+analytic Cox--Snell + 1 Newton & 3.91e-4 & 2.516 & 0.951 & 1.95e-4 & 2.545 & 0.947 & 9.82e-5 & 2.511 & 0.949 \\
19+CSL, 1 round & 1.249 & 2.791 & -- & 0.432 & 2.586 & -- & 0.152 & 2.519 & -- \\
20+CSL, 2 rounds & 0.494 & 2.563 & -- & 0.110 & 2.545 & -- & 0.026 & 2.511 & -- \\
21+AEE (Lin--Xi) & 6.200 & 6.700 & -- & 3.072 & 3.992 & -- & 1.529 & 2.952 & -- \\
22+(2,1) stale Hessian & 0.003 & 2.516 & -- & 0.001 & 2.545 & -- & 4.32e-4 & 2.511 & -- \\
23+centralized MLE & 0 & 2.516 & 0.951 & 0 & 2.545 & 0.947 & 0 & 2.511 & 0.949 \\
24+\bottomrule
25+\end{tabular}
26+\begin{tablenotes}\footnotesize
27+\item 1000 paired replications per column. gap $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$; RMSE $=$ root mean square of $\sqrt N\,\lVert\tilde\theta - \theta_0\rVert$; cov $=$ empirical coverage of nominal 95\% Wald intervals (observed information at the estimate; `--' when not computed). Monte Carlo standard errors of the gaps and RMSEs are in \texttt{results/processed/summary.csv}; coverage standard error $\approx 0.007$.
28+\item Fraction of workers with all local fits valid (orders 2 / 3): $m=256$: 1.000 / 1.000; $m=1024$: 1.000 / 1.000; $m=512$: 1.000 / 1.000.
29+\end{tablenotes}
30+\end{threeparttable}
31+\end{table}
added paper/tables/mc_rates.tex +18 −0
@@ -0,0 +1,18 @@
1+\begin{table}[t]
2+\centering\small
3+\begin{threeparttable}
4+\caption{Empirical log-log slopes of the RMS scaled gap to $\widehat\theta_N$ against $m$ at fixed $N$ (largest $N$ of each design), with the slopes predicted by the leading bias term.}\label{tab:mc_rates}
5+\begin{tabular}{@{}llrrrrrrrrr@{}}
6+\toprule
7+Design & $p$ & $N$ & r1\_q0 & r1\_q1 & r1\_q2 & r2\_q0 & r2\_q1 & r3\_q0 & r3\_q1 & cs\_q1 \\
8+predicted & & & -1 & -2 & -4 & -2 & -4 & -3 & -6 & -4 \\
9+\midrule
10+exp & 1 & 262,144 & -1.02 & -2.04 & -4.10 & -1.18 & -2.24 & -0.98 & -1.87 & -1.04 \\
11+pois & 8 & 262,144 & -1.00 & -1.98 & -3.87 & -0.61 & -1.15 & -0.54 & -1.05 & -1.00 \\
12+\bottomrule
13+\end{tabular}
14+\begin{tablenotes}\footnotesize
15+\item Slopes are least-squares fits over the available $m$ grid; they approach the predicted exponent only where the bias term dominates the random remainder and the sample is in the asymptotic regime. The predicted slope for the Cox--Snell + 1 Newton estimator assumes an exact first-order bias formula ($-4$). Source: \texttt{results/processed/summary.csv}.
16+\end{tablenotes}
17+\end{threeparttable}
18+\end{table}
modified paper/tables/numbers.tex +4 −2
@@ -1,4 +1,6 @@
1−% Auto-generated numeric macros (experiments/make_tables.py). Placeholder values are overwritten at build time. 1+% Auto-generated by experiments/make_tables.py
2+\newcommand{\numMCcells}{49}
3+\newcommand{\numManifest}{49}
2 \newcommand{\numMCrepsExp}{2000} 4 \newcommand{\numMCrepsExp}{2000}
3 \newcommand{\numMCrepsLogit}{1000} 5 \newcommand{\numMCrepsLogit}{1000}
4−\newcommand{\benchHosts}{16} 6+\newcommand{\benchHosts}{0}
added paper/tables/skew.tex +51 −0
@@ -0,0 +1,51 @@
1+\begin{table}[t]
2+\centering\scriptsize
3+\begin{threeparttable}
4+\caption{Unequal block sizes (Proposition~\ref{prop:robust}): RMS of $\sqrt N\,\lVert\tilde\theta - \widehat\theta_N\rVert$ for size-weighted and equal-weight aggregation of the raw and half-block-jackknife local estimates, with $q = 0, 1, 2$ Newton corrections. Sizes grow geometrically with max/min ratio as shown; $N \approx 2^{18}$.}\label{tab:skew}
5+\begin{tabular}{@{}llrrllrrr@{}}
6+\toprule
7+Model & $K$ & ratio & reps & weights & pilot & $q=0$ & $q=1$ & $q=2$ \\
8+\midrule
9+exponential & 256 & 1 & 1000 & equal & jack2 & 0.062 & 1.24e-5 & 8.52e-13 \\
10+exponential & 256 & 1 & 1000 & equal & raw & 0.501 & 4.97e-4 & 5.13e-10 \\
11+exponential & 256 & 1 & 1000 & weighted & jack2 & 0.062 & 1.24e-5 & 8.52e-13 \\
12+exponential & 256 & 1 & 1000 & weighted & raw & 0.501 & 4.97e-4 & 5.13e-10 \\
13+exponential & 256 & 4 & 1000 & equal & jack2 & 0.441 & 6.71e-4 & 2.96e-9 \\
14+exponential & 256 & 4 & 1000 & equal & raw & 0.735 & 0.002 & 1.03e-8 \\
15+exponential & 256 & 4 & 1000 & weighted & jack2 & 0.064 & 1.37e-5 & 1.13e-12 \\
16+exponential & 256 & 4 & 1000 & weighted & raw & 0.503 & 5.02e-4 & 5.25e-10 \\
17+exponential & 256 & 16 & 1000 & equal & jack2 & 0.943 & 0.003 & 5.96e-8 \\
18+exponential & 256 & 16 & 1000 & equal & raw & 1.321 & 0.006 & 1.78e-7 \\
19+exponential & 256 & 16 & 1000 & weighted & jack2 & 0.062 & 1.28e-5 & 1.05e-12 \\
20+exponential & 256 & 16 & 1000 & weighted & raw & 0.500 & 4.95e-4 & 5.07e-10 \\
21+exponential & 1024 & 4 & 1000 & equal & jack2 & 0.447 & 6.83e-4 & 3.06e-9 \\
22+exponential & 1024 & 4 & 1000 & equal & raw & 2.415 & 0.012 & 3.50e-7 \\
23+exponential & 1024 & 4 & 1000 & weighted & jack2 & 0.123 & 5.26e-5 & 2.04e-11 \\
24+exponential & 1024 & 4 & 1000 & weighted & raw & 2.009 & 0.008 & 1.24e-7 \\
25+exponential & 1024 & 16 & 1000 & equal & jack2 & 0.994 & 0.003 & 6.41e-8 \\
26+exponential & 1024 & 16 & 1000 & equal & raw & 3.917 & 0.033 & 3.05e-6 \\
27+exponential & 1024 & 16 & 1000 & weighted & jack2 & 0.133 & 5.70e-5 & 1.83e-11 \\
28+exponential & 1024 & 16 & 1000 & weighted & raw & 2.012 & 0.008 & 1.25e-7 \\
29+logistic & 256 & 1 & 300 & equal & jack2 & 0.508 & 2.85e-4 & 1.33e-9 \\
30+logistic & 256 & 1 & 300 & equal & raw & 4.034 & 0.014 & 1.77e-7 \\
31+logistic & 256 & 1 & 300 & weighted & jack2 & 0.508 & 2.85e-4 & 1.33e-9 \\
32+logistic & 256 & 1 & 300 & weighted & raw & 4.034 & 0.014 & 1.77e-7 \\
33+logistic & 256 & 4 & 300 & equal & jack2 & 2.025 & 0.004 & 3.67e-8 \\
34+logistic & 256 & 4 & 300 & equal & raw & 5.096 & 0.025 & 8.06e-7 \\
35+logistic & 256 & 4 & 300 & weighted & jack2 & 0.522 & 3.09e-4 & 1.35e-9 \\
36+logistic & 256 & 4 & 300 & weighted & raw & 3.998 & 0.014 & 1.70e-7 \\
37+logistic & 256 & 16 & 300 & equal & jack2 & 4.383 & 0.020 & 7.36e-7 \\
38+logistic & 256 & 16 & 300 & equal & raw & 8.944 & 0.083 & 1.02e-5 \\
39+logistic & 256 & 16 & 300 & weighted & jack2 & 0.526 & 3.29e-4 & 1.36e-9 \\
40+logistic & 256 & 16 & 300 & weighted & raw & 4.061 & 0.014 & 1.84e-7 \\
41+logistic & 1024 & 4 & 240 & equal & jack2 & 3.625 & 0.015 & 3.66e-7 \\
42+logistic & 1024 & 4 & 240 & equal & raw & 19.769 & 0.344 & 1.07e-4 \\
43+logistic & 1024 & 4 & 240 & weighted & jack2 & 2.138 & 0.005 & 4.30e-8 \\
44+logistic & 1024 & 4 & 240 & weighted & raw & 16.616 & 0.240 & 5.00e-5 \\
45+\bottomrule
46+\end{tabular}
47+\begin{tablenotes}\footnotesize
48+\item Source: \texttt{results/processed/skew\_summary.csv}.
49+\end{tablenotes}
50+\end{threeparttable}
51+\end{table}
modified results/figures/fig_boundary_exponential.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_exp_contam_p1_N1048576.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_exp_p1_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_logit_misspec_p3_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_logit_p3_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_logit_p8_N16384.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_pois_p3_N65536.pdf +0 −0

Binary file not shown.

added results/figures/fig_coverage95_pois_p8_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_gap_collapse_exp_contam_p1.pdf +0 −0

Binary file not shown.

modified results/figures/fig_gap_collapse_exp_p1.pdf +0 −0

Binary file not shown.

added results/figures/fig_gap_collapse_logit_misspec_p3.pdf +0 −0

Binary file not shown.

added results/figures/fig_gap_collapse_logit_p3.pdf +0 −0

Binary file not shown.

added results/figures/fig_gap_collapse_logit_p8.pdf +0 −0

Binary file not shown.

added results/figures/fig_gap_collapse_pois_p3.pdf +0 −0

Binary file not shown.

added results/figures/fig_gap_collapse_pois_p8.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_exp_contam_p1_N1048576.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_exp_p1_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_logit_misspec_p3_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_logit_p3_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_logit_p8_N16384.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_pois_p3_N65536.pdf +0 −0

Binary file not shown.

added results/figures/fig_rms_gap_vs_m_pois_p8_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_exp_contam_p1_N1048576.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_exp_p1_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_logit_misspec_p3_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_logit_p3_N262144.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_logit_p8_N16384.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_pois_p3_N65536.pdf +0 −0

Binary file not shown.

added results/figures/fig_rmse_theta0_vs_m_pois_p8_N262144.pdf +0 −0

Binary file not shown.

added results/processed/ablation_exponential_summary.csv +505 −0
@@ -0,0 +1,505 @@
1+N,m,variant,pattern,r,q,hessian,bits,rms,mean0,n,valid
2+196608,12,pat1-2-3-4_q0,1-2-3-4,4,0,exact,64,5.038998371871043,-1.3496118489731237,1000,1.0
3+196608,12,pat1-2-3-4_q1,1-2-3-4,4,1,exact,64,0.09916665324815943,-0.05726758970955714,1000,1.0
4+196608,12,pat1-2-3-4_q2,1-2-3-4,4,2,exact,64,7.137598388445489e-05,-2.2180113475051842e-05,1000,1.0
5+196608,12,pat1-2-3-4_q3,1-2-3-4,4,3,exact,64,1.005558735439718e-10,-1.1487317577410676e-11,1000,1.0
6+196608,12,pat1-2-3_q0,1-2-3,3,0,exact,64,3.2572097018805266,2.68150568149954,1000,1.0
7+196608,12,pat1-2-3_q1,1-2-3,3,1,exact,64,0.034394708433008125,-0.023922398225239686,1000,1.0
8+196608,12,pat1-2-3_q2,1-2-3,3,2,exact,64,5.9173395734414595e-06,-2.6669741319861644e-06,1000,1.0
9+196608,12,pat1-2-3_q3,1-2-3,3,3,exact,64,3.6227479710205474e-13,-7.723848844635678e-14,1000,1.0
10+196608,12,pat1-2-4_q0,1-2-4,3,0,exact,64,4.215558987006441,4.025211524990359,1000,1.0
11+196608,12,pat1-2-4_q1,1-2-4,3,1,exact,64,0.04614755979095163,-0.04007532784275546,1000,1.0
12+196608,12,pat1-2-4_q1_Hbits12,1-2-4,3,1,quantizedH,12,0.046138313265687396,-0.04006664685851366,1000,
13+196608,12,pat1-2-4_q1_Hbits16,1-2-4,3,1,quantizedH,16,0.046147187432413875,-0.04007484871152553,1000,
14+196608,12,pat1-2-4_q1_Hbits2,1-2-4,3,1,quantizedH,2,0.050419496488955017,0.03984627578488361,1000,
15+196608,12,pat1-2-4_q1_Hbits24,1-2-4,3,1,quantizedH,24,0.04614755666362255,-0.04007532586063331,1000,
16+196608,12,pat1-2-4_q1_Hbits3,1-2-4,3,1,quantizedH,3,0.050419496488955017,0.03984627578488361,1000,
17+196608,12,pat1-2-4_q1_Hbits4,1-2-4,3,1,quantizedH,4,0.07089518639239879,0.023840079038142833,1000,
18+196608,12,pat1-2-4_q1_Hbits52,1-2-4,3,1,quantizedH,52,0.046147559790951406,-0.04007532784275521,1000,
19+196608,12,pat1-2-4_q1_Hbits6,1-2-4,3,1,quantizedH,6,0.04960686773239981,-0.039453237639831065,1000,
20+196608,12,pat1-2-4_q1_Hbits8,1-2-4,3,1,quantizedH,8,0.046325203614386315,-0.039952212562446986,1000,
21+196608,12,pat1-2-4_q1_bothbits12,1-2-4,3,1,quantizedBoth,12,0.04613457187806652,-0.04005302645917944,1000,
22+196608,12,pat1-2-4_q1_bothbits16,1-2-4,3,1,quantizedBoth,16,0.04614691755652101,-0.04007446161046857,1000,
23+196608,12,pat1-2-4_q1_bothbits2,1-2-4,3,1,quantizedBoth,2,0.4502167314850321,0.07956257331938836,1000,
24+196608,12,pat1-2-4_q1_bothbits24,1-2-4,3,1,quantizedBoth,24,0.046147556905197014,-0.04007532723766722,1000,
25+196608,12,pat1-2-4_q1_bothbits3,1-2-4,3,1,quantizedBoth,3,0.22700738023529587,0.04427203811517248,1000,
26+196608,12,pat1-2-4_q1_bothbits4,1-2-4,3,1,quantizedBoth,4,0.13175189352425834,0.02838661757871094,1000,
27+196608,12,pat1-2-4_q1_bothbits52,1-2-4,3,1,quantizedBoth,52,0.04614755979095144,-0.040075327842755255,1000,
28+196608,12,pat1-2-4_q1_bothbits6,1-2-4,3,1,quantizedBoth,6,0.057540317571334196,-0.03958981599109757,1000,
29+196608,12,pat1-2-4_q1_bothbits8,1-2-4,3,1,quantizedBoth,8,0.046794728249255214,-0.03987071941166932,1000,
30+196608,12,pat1-2-4_q1_gbits12,1-2-4,3,1,quantizedg,12,0.046143855323002476,-0.04006170757805265,1000,
31+196608,12,pat1-2-4_q1_gbits16,1-2-4,3,1,quantizedg,16,0.046147290270765834,-0.040074940742832216,1000,
32+196608,12,pat1-2-4_q1_gbits2,1-2-4,3,1,quantizedg,2,0.4520641212711294,0.00039960342995543053,1000,
33+196608,12,pat1-2-4_q1_gbits24,1-2-4,3,1,quantizedg,24,0.04614756003252581,-0.04007532921978898,1000,
34+196608,12,pat1-2-4_q1_gbits3,1-2-4,3,1,quantizedg,3,0.22956356232707814,-0.03558313513809876,1000,
35+196608,12,pat1-2-4_q1_gbits4,1-2-4,3,1,quantizedg,4,0.11719932981292415,-0.035412355998374834,1000,
36+196608,12,pat1-2-4_q1_gbits52,1-2-4,3,1,quantizedg,52,0.04614755979095163,-0.04007532784275546,1000,
37+196608,12,pat1-2-4_q1_gbits6,1-2-4,3,1,quantizedg,6,0.054633122680685084,-0.04021169198144978,1000,
38+196608,12,pat1-2-4_q1_gbits8,1-2-4,3,1,quantizedg,8,0.04663100668110985,-0.039994028035955406,1000,
39+196608,12,pat1-2-4_q1_single,1-2-4,3,1,single,64,0.04614755979095136,-0.040075327842755304,1000,
40+196608,12,pat1-2-4_q1_stale,1-2-4,3,1,stale,64,0.36577346056184734,0.3463871401561131,1000,
41+196608,12,pat1-2-4_q2,1-2-4,3,2,exact,64,7.120861782529493e-06,-4.802108244655432e-06,1000,1.0
42+196608,12,pat1-2-4_q2_single,1-2-4,3,2,single,64,7.120861782468348e-06,-4.802108244458522e-06,1000,
43+196608,12,pat1-2-4_q2_stale,1-2-4,3,2,stale,64,0.028495627470691943,0.026940899370413243,1000,
44+196608,12,pat1-2-4_q3,1-2-4,3,3,exact,64,2.9406787809745057e-13,-1.1371632142197843e-13,1000,1.0
45+196608,12,pat1-2-4_q3_single,1-2-4,3,3,single,64,2.941337978198875e-13,-1.1401168849060695e-13,1000,
46+196608,12,pat1-2-4_q3_stale,1-2-4,3,3,stale,64,0.00219756053799601,0.0020756576599389055,1000,
47+196608,12,pat1-2_q0,1-2,2,0,exact,64,8.10831292733407,-8.071193594172236,1000,1.0
48+196608,12,pat1-2_q1,1-2,2,1,exact,64,0.15098814216125261,-0.14828187438015467,1000,1.0
49+196608,12,pat1-2_q1_Hbits12,1-2,2,1,quantizedH,12,0.15096808064934975,-0.1482608798950797,1000,
50+196608,12,pat1-2_q1_Hbits16,1-2,2,1,quantizedH,16,0.15098854692605213,-0.14828227778697217,1000,
51+196608,12,pat1-2_q1_Hbits2,1-2,2,1,quantizedH,2,0.1598255053862907,0.15199148086592215,1000,
52+196608,12,pat1-2_q1_Hbits24,1-2,2,1,quantizedH,24,0.15098814806673513,-0.1482818810166552,1000,
53+196608,12,pat1-2_q1_Hbits3,1-2,2,1,quantizedH,3,0.1598255053862907,0.15199148086592215,1000,
54+196608,12,pat1-2_q1_Hbits4,1-2,2,1,quantizedH,4,0.1598255053862907,0.15199148086592215,1000,
55+196608,12,pat1-2_q1_Hbits52,1-2,2,1,quantizedH,52,0.15098814216125256,-0.14828187438015467,1000,
56+196608,12,pat1-2_q1_Hbits6,1-2,2,1,quantizedH,6,0.1314402834998777,-0.11671341879293311,1000,
57+196608,12,pat1-2_q1_Hbits8,1-2,2,1,quantizedH,8,0.15139270366359195,-0.14777559268090582,1000,
58+196608,12,pat1-2_q1_bothbits12,1-2,2,1,quantizedBoth,12,0.15096213989450954,-0.14824594783444842,1000,
59+196608,12,pat1-2_q1_bothbits16,1-2,2,1,quantizedBoth,16,0.1509882221546252,-0.14828174422817123,1000,
60+196608,12,pat1-2_q1_bothbits2,1-2,2,1,quantizedBoth,2,1.1023460355617418,-0.20075472457925772,1000,
61+196608,12,pat1-2_q1_bothbits24,1-2,2,1,quantizedBoth,24,0.15098813882561912,-0.14828187204885404,1000,
62+196608,12,pat1-2_q1_bothbits3,1-2,2,1,quantizedBoth,3,0.5351968813354548,0.17163619904805094,1000,
63+196608,12,pat1-2_q1_bothbits4,1-2,2,1,quantizedBoth,4,0.29783112168735,0.15604774177993075,1000,
64+196608,12,pat1-2_q1_bothbits52,1-2,2,1,quantizedBoth,52,0.15098814216125273,-0.14828187438015475,1000,
65+196608,12,pat1-2_q1_bothbits6,1-2,2,1,quantizedBoth,6,0.1432945262217917,-0.11520191196297348,1000,
66+196608,12,pat1-2_q1_bothbits8,1-2,2,1,quantizedBoth,8,0.15191028240968446,-0.14757289264332235,1000,
67+196608,12,pat1-2_q1_gbits12,1-2,2,1,quantizedg,12,0.1509820116352656,-0.14826694585569267,1000,
68+196608,12,pat1-2_q1_gbits16,1-2,2,1,quantizedg,16,0.1509878169030937,-0.14828134083152228,1000,
69+196608,12,pat1-2_q1_gbits2,1-2,2,1,quantizedg,2,1.1311752812698943,-0.4897125055392436,1000,
70+196608,12,pat1-2_q1_gbits24,1-2,2,1,quantizedg,24,0.15098813292014338,-0.1482818654123503,1000,
71+196608,12,pat1-2_q1_gbits3,1-2,2,1,quantizedg,3,0.5032469272389815,-0.12938415346299964,1000,
72+196608,12,pat1-2_q1_gbits4,1-2,2,1,quantizedg,4,0.2769546540312099,-0.14443934535423142,1000,
73+196608,12,pat1-2_q1_gbits52,1-2,2,1,quantizedg,52,0.15098814216125267,-0.14828187438015467,1000,
74+196608,12,pat1-2_q1_gbits6,1-2,2,1,quantizedg,6,0.16054593320017918,-0.1467867504620175,1000,
75+196608,12,pat1-2_q1_gbits8,1-2,2,1,quantizedg,8,0.15153602500764246,-0.14807869642073734,1000,
76+196608,12,pat1-2_q1_single,1-2,2,1,single,64,0.15098814216125264,-0.14828187438015472,1000,
77+196608,12,pat1-2_q1_stale,1-2,2,1,stale,64,0.48182346774593626,-0.48087647653878507,1000,
78+196608,12,pat1-2_q2,1-2,2,2,exact,64,5.517120345426267e-05,-5.141950996465999e-05,1000,1.0
79+196608,12,pat1-2_q2_single,1-2,2,2,single,64,5.5171203454369e-05,-5.141950996465999e-05,1000,
80+196608,12,pat1-2_q2_stale,1-2,2,2,stale,64,0.03655154047163949,-0.036486415709318,1000,
81+196608,12,pat1-2_q3,1-2,2,3,exact,64,8.976220793355467e-12,-6.863099978807412e-12,1000,1.0
82+196608,12,pat1-2_q3_single,1-2,2,3,single,64,8.976303946145025e-12,-6.863198434496955e-12,1000,
83+196608,12,pat1-2_q3_stale,1-2,2,3,stale,64,0.0028075531160814625,-0.0028026362219081974,1000,
84+196608,12,pat1-3_q0,1-3,2,0,exact,64,13.466334067690298,-13.4475432320081,1000,1.0
85+196608,12,pat1-3_q1,1-3,2,1,exact,64,0.4112537675940219,-0.4089880704235383,1000,1.0
86+196608,12,pat1-3_q2,1-3,2,2,exact,64,0.0003898181684734085,-0.00038144433561693245,1000,1.0
87+196608,12,pat1-3_q3,1-3,2,3,exact,64,3.729168853570627e-10,-3.42714065134753e-10,1000,1.0
88+196608,12,pat1_q0,1,1,0,exact,64,40.295051090314445,40.292216745925636,1000,1.0
89+196608,12,pat1_q1,1,1,1,exact,64,3.6629928065934423,-3.661986406318907,1000,1.0
90+196608,12,pat1_q1_Hbits12,1,1,1,quantizedH,12,3.6629076948523984,-3.6619016865212357,1000,
91+196608,12,pat1_q1_Hbits16,1,1,1,quantizedH,16,3.66299000064699,-3.6619834010125865,1000,
92+196608,12,pat1_q1_Hbits2,1,1,1,quantizedH,2,8.961390947888326,-8.958791667138366,1000,
93+196608,12,pat1_q1_Hbits24,1,1,1,quantizedH,24,3.6629927727448646,-3.6619863723091894,1000,
94+196608,12,pat1_q1_Hbits3,1,1,1,quantizedH,3,1.9323849784322313,-1.922933322414932,1000,
95+196608,12,pat1_q1_Hbits4,1,1,1,quantizedH,4,4.6553607552300065,-4.496961530455964,1000,
96+196608,12,pat1_q1_Hbits52,1,1,1,quantizedH,52,3.6629928065934427,-3.6619864063189076,1000,
97+196608,12,pat1_q1_Hbits6,1,1,1,quantizedH,6,3.6152201128040695,-3.6075911805112524,1000,
98+196608,12,pat1_q1_Hbits8,1,1,1,quantizedH,8,3.664473390754607,-3.663002442867629,1000,
99+196608,12,pat1_q1_bothbits12,1,1,1,quantizedBoth,12,3.6630381114163444,-3.6620312857247583,1000,
100+196608,12,pat1_q1_bothbits16,1,1,1,quantizedBoth,16,3.6629934124643113,-3.661986930476007,1000,
101+196608,12,pat1_q1_bothbits2,1,1,1,quantizedBoth,2,15.140953832098015,-15.133409096278431,1000,
102+196608,12,pat1_q1_bothbits24,1,1,1,quantizedBoth,24,3.6629927697800797,-3.661986367695426,1000,
103+196608,12,pat1_q1_bothbits3,1,1,1,quantizedBoth,3,0.9020140753488401,0.6787301618475311,1000,
104+196608,12,pat1_q1_bothbits4,1,1,1,quantizedBoth,4,5.773200593850328,-5.539294170274072,1000,
105+196608,12,pat1_q1_bothbits52,1,1,1,quantizedBoth,52,3.6629928065934427,-3.6619864063189076,1000,
106+196608,12,pat1_q1_bothbits6,1,1,1,quantizedBoth,6,3.634929259101741,-3.6151760242200326,1000,
107+196608,12,pat1_q1_bothbits8,1,1,1,quantizedBoth,8,3.663120355632819,-3.660815292715375,1000,
108+196608,12,pat1_q1_gbits12,1,1,1,quantizedg,12,3.6631230061477327,-3.662115989439731,1000,
109+196608,12,pat1_q1_gbits16,1,1,1,quantizedg,16,3.6629962187976086,-3.6619899358126706,1000,
110+196608,12,pat1_q1_gbits2,1,1,1,quantizedg,2,9.180894625300507,-9.172093971774435,1000,
111+196608,12,pat1_q1_gbits24,1,1,1,quantizedg,24,3.6629928036286725,-3.661986401705144,1000,
112+196608,12,pat1_q1_gbits3,1,1,1,quantizedg,3,1.1038431100683848,-0.9528624374220762,1000,
113+196608,12,pat1_q1_gbits4,1,1,1,quantizedg,4,4.789042554721146,-4.683737584681572,1000,
114+196608,12,pat1_q1_gbits52,1,1,1,quantizedg,52,3.6629928065934423,-3.661986406318907,1000,
115+196608,12,pat1_q1_gbits6,1,1,1,quantizedg,6,3.682977610999207,-3.669609951738764,1000,
116+196608,12,pat1_q1_gbits8,1,1,1,quantizedg,8,3.6616714652507647,-3.659801723594189,1000,
117+196608,12,pat1_q1_single,1,1,1,single,64,3.6629928065934423,-3.661986406318907,1000,
118+196608,12,pat1_q1_stale,1,1,1,stale,64,6.198698426211689,6.19729525327051,1000,
119+196608,12,pat1_q2,1,1,2,exact,64,0.030294310357980872,-0.030261193522678025,1000,1.0
120+196608,12,pat1_q2_single,1,1,2,single,64,0.03029431035798004,-0.03026119352267714,1000,
121+196608,12,pat1_q2_stale,1,1,2,stale,64,0.5558186917192648,0.5555289075944222,1000,
122+196608,12,pat1_q3,1,1,3,exact,64,2.07890901269474e-06,-2.0698423692765244e-06,1000,1.0
123+196608,12,pat1_q3_single,1,1,3,single,64,2.0789090126839314e-06,-2.0698423692765244e-06,1000,
124+196608,12,pat1_q3_stale,1,1,3,stale,64,0.04341264486664452,0.043373544721287696,1000,
125+393216,24,pat1-2-3-4_q0,1-2-3-4,4,0,exact,64,2.408101248232987,0.02145241315256412,1000,1.0
126+393216,24,pat1-2-3-4_q1,1-2-3-4,4,1,exact,64,0.01629908806181326,-0.009247365576410944,1000,1.0
127+393216,24,pat1-2-3-4_q2,1-2-3-4,4,2,exact,64,1.477030688434044e-06,-4.2356125067304486e-07,1000,1.0
128+393216,24,pat1-2-3-4_q3,1-2-3-4,4,3,exact,64,7.992516204182178e-14,-3.620171657551204e-15,1000,1.0
129+393216,24,pat1-2-3_q0,1-2-3,3,0,exact,64,1.0715165585861564,0.3642844134731204,1000,1.0
130+393216,24,pat1-2-3_q1,1-2-3,3,1,exact,64,0.003190363760616121,-0.0018308793341250717,1000,1.0
131+393216,24,pat1-2-3_q2,1-2-3,3,2,exact,64,5.11934776578555e-08,-1.6229168833308095e-08,1000,1.0
132+393216,24,pat1-2-3_q3,1-2-3,3,3,exact,64,7.390737513733653e-14,1.4619924001649095e-15,1000,1.0
133+393216,24,pat1-2-4-8_q0,1-2-4-8,4,0,exact,64,0.9275494372622984,-0.2733759512871766,1000,1.0
134+393216,24,pat1-2-4-8_q1,1-2-4-8,4,1,exact,64,0.002431862685393414,-0.001372025795885134,1000,1.0
135+393216,24,pat1-2-4-8_q2,1-2-4-8,4,2,exact,64,3.144089422126544e-08,-9.431290347390969e-09,1000,1.0
136+393216,24,pat1-2-4-8_q3,1-2-4-8,4,3,exact,64,7.42671830293128e-14,-9.746616001099397e-16,1000,1.0
137+393216,24,pat1-2-4_q0,1-2-4,3,0,exact,64,0.8462636943929642,0.47856174691320724,1000,1.0
138+393216,24,pat1-2-4_q1,1-2-4,3,1,exact,64,0.001850604300724433,-0.0011420293801571728,1000,1.0
139+393216,24,pat1-2-4_q1_Hbits12,1-2-4,3,1,quantizedH,12,0.0018545752836736397,-0.0011433904890474586,1000,
140+393216,24,pat1-2-4_q1_Hbits16,1-2-4,3,1,quantizedH,16,0.001850463401880276,-0.0011421974617886982,1000,
141+393216,24,pat1-2-4_q1_Hbits2,1-2-4,3,1,quantizedH,2,0.003401474830143217,0.0012128928755874098,1000,
142+393216,24,pat1-2-4_q1_Hbits24,1-2-4,3,1,quantizedH,24,0.0018506036075663203,-0.0011420287493048775,1000,
143+393216,24,pat1-2-4_q1_Hbits3,1-2-4,3,1,quantizedH,3,0.003401474830143217,0.0012128928755874098,1000,
144+393216,24,pat1-2-4_q1_Hbits4,1-2-4,3,1,quantizedH,4,0.003401474830143217,0.0012128928755874098,1000,
145+393216,24,pat1-2-4_q1_Hbits52,1-2-4,3,1,quantizedH,52,0.001850604300724433,-0.0011420293801571728,1000,
146+393216,24,pat1-2-4_q1_Hbits6,1-2-4,3,1,quantizedH,6,0.004060004768684619,4.786207168444577e-06,1000,
147+393216,24,pat1-2-4_q1_Hbits8,1-2-4,3,1,quantizedH,8,0.002215450107241173,-0.0011775308014106132,1000,
148+393216,24,pat1-2-4_q1_bothbits12,1-2-4,3,1,quantizedBoth,12,0.0018571135620659934,-0.001143655740699415,1000,
149+393216,24,pat1-2-4_q1_bothbits16,1-2-4,3,1,quantizedBoth,16,0.00185043366270375,-0.0011423613847107174,1000,
150+393216,24,pat1-2-4_q1_bothbits2,1-2-4,3,1,quantizedBoth,2,0.08993867834112033,0.006823870382718283,1000,
151+393216,24,pat1-2-4_q1_bothbits24,1-2-4,3,1,quantizedBoth,24,0.0018506033023150035,-0.0011420279145724506,1000,
152+393216,24,pat1-2-4_q1_bothbits3,1-2-4,3,1,quantizedBoth,3,0.04508044119881075,0.00259826136843461,1000,
153+393216,24,pat1-2-4_q1_bothbits4,1-2-4,3,1,quantizedBoth,4,0.02231112874082009,0.00011274092325826683,1000,
154+393216,24,pat1-2-4_q1_bothbits52,1-2-4,3,1,quantizedBoth,52,0.0018506043007246163,-0.0011420293801573122,1000,
155+393216,24,pat1-2-4_q1_bothbits6,1-2-4,3,1,quantizedBoth,6,0.006605273186282643,-0.00020570529697675968,1000,
156+393216,24,pat1-2-4_q1_bothbits8,1-2-4,3,1,quantizedBoth,8,0.0026593390103624988,-0.0012279490281015448,1000,
157+393216,24,pat1-2-4_q1_gbits12,1-2-4,3,1,quantizedg,12,0.0018530066686658127,-0.001142294488347924,1000,
158+393216,24,pat1-2-4_q1_gbits16,1-2-4,3,1,quantizedg,16,0.0018505752222783645,-0.0011421933036023751,1000,
159+393216,24,pat1-2-4_q1_gbits2,1-2-4,3,1,quantizedg,2,0.09008949753313411,0.0044915488246097856,1000,
160+393216,24,pat1-2-4_q1_gbits24,1-2-4,3,1,quantizedg,24,0.0018506039954657748,-0.0011420285454255812,1000,
161+393216,24,pat1-2-4_q1_gbits3,1-2-4,3,1,quantizedg,3,0.044829536124334164,0.0002500332233302373,1000,
162+393216,24,pat1-2-4_q1_gbits4,1-2-4,3,1,quantizedg,4,0.022292588718330997,-0.0022452098191569647,1000,
163+393216,24,pat1-2-4_q1_gbits52,1-2-4,3,1,quantizedg,52,0.001850604300724433,-0.0011420293801571728,1000,
164+393216,24,pat1-2-4_q1_gbits6,1-2-4,3,1,quantizedg,6,0.0059215607907765,-0.0013536478187344005,1000,
165+393216,24,pat1-2-4_q1_gbits8,1-2-4,3,1,quantizedg,8,0.0023348855471433654,-0.0011923747975537628,1000,
166+393216,24,pat1-2-4_q1_single,1-2-4,3,1,single,64,0.001850604300724433,-0.0011420293801571728,1000,
167+393216,24,pat1-2-4_q1_stale,1-2-4,3,1,stale,64,0.03537453240487412,0.02035139140625425,1000,
168+393216,24,pat1-2-4_q2,1-2-4,3,2,exact,64,1.539498404584012e-08,-5.460963643088784e-09,1000,1.0
169+393216,24,pat1-2-4_q2_single,1-2-4,3,2,single,64,1.539498404584012e-08,-5.460963643088784e-09,1000,
170+393216,24,pat1-2-4_q2_stale,1-2-4,3,2,stale,64,0.0014234045324019833,0.000820796342651219,1000,
171+393216,24,pat1-2-4_q3,1-2-4,3,3,exact,64,7.198046914541243e-14,3.480934286106927e-16,1000,1.0
172+393216,24,pat1-2-4_q3_single,1-2-4,3,3,single,64,7.198046914541243e-14,3.480934286106927e-16,1000,
173+393216,24,pat1-2-4_q3_stale,1-2-4,3,3,stale,64,5.71877733182497e-05,3.3032149150550144e-05,1000,
174+393216,24,pat1-2_q0,1-2,2,0,exact,64,2.52744143063154,-2.4844268874507778,1000,1.0
175+393216,24,pat1-2_q1,1-2,2,1,exact,64,0.010848877501410942,-0.010186603223624779,1000,1.0
176+393216,24,pat1-2_q1_Hbits12,1-2,2,1,quantizedH,12,0.010822239661321694,-0.010154697657611293,1000,
177+393216,24,pat1-2_q1_Hbits16,1-2,2,1,quantizedH,16,0.010848245902675689,-0.010186303836556802,1000,
178+393216,24,pat1-2_q1_Hbits2,1-2,2,1,quantizedH,2,0.013532975374507919,0.010062038113977363,1000,
179+393216,24,pat1-2_q1_Hbits24,1-2,2,1,quantizedH,24,0.01084887644098738,-0.010186603223238532,1000,
180+393216,24,pat1-2_q1_Hbits3,1-2,2,1,quantizedH,3,0.013532975374507919,0.010062038113977363,1000,
181+393216,24,pat1-2_q1_Hbits4,1-2,2,1,quantizedH,4,0.013532975374507919,0.010062038113977363,1000,
182+393216,24,pat1-2_q1_Hbits52,1-2,2,1,quantizedH,52,0.01084887750141015,-0.010186603223624151,1000,
183+393216,24,pat1-2_q1_Hbits6,1-2,2,1,quantizedH,6,0.014677935176906026,0.008505760293443809,1000,
184+393216,24,pat1-2_q1_Hbits8,1-2,2,1,quantizedH,8,0.01212310956103166,-0.010072263516237516,1000,
185+393216,24,pat1-2_q1_bothbits12,1-2,2,1,quantizedBoth,12,0.010834473032905578,-0.010165111872143766,1000,
186+393216,24,pat1-2_q1_bothbits16,1-2,2,1,quantizedBoth,16,0.01084872388735015,-0.010186851914999831,1000,
187+393216,24,pat1-2_q1_bothbits2,1-2,2,1,quantizedBoth,2,0.28388601450655143,-0.044735103638732274,1000,
188+393216,24,pat1-2_q1_bothbits24,1-2,2,1,quantizedBoth,24,0.010848873039795042,-0.010186601043121175,1000,
189+393216,24,pat1-2_q1_bothbits3,1-2,2,1,quantizedBoth,3,0.14048553576088477,0.0031830394544638454,1000,
190+393216,24,pat1-2_q1_bothbits4,1-2,2,1,quantizedBoth,4,0.07074072663357625,0.008388205157877887,1000,
191+393216,24,pat1-2_q1_bothbits52,1-2,2,1,quantizedBoth,52,0.010848877501410163,-0.010186603223624151,1000,
192+393216,24,pat1-2_q1_bothbits6,1-2,2,1,quantizedBoth,6,0.022457994899373467,0.007633177779776187,1000,
193+393216,24,pat1-2_q1_bothbits8,1-2,2,1,quantizedBoth,8,0.013098890071383764,-0.01034205174878799,1000,
194+393216,24,pat1-2_q1_gbits12,1-2,2,1,quantizedg,12,0.010860815823815033,-0.01019701870793441,1000,
195+393216,24,pat1-2_q1_gbits16,1-2,2,1,quantizedg,16,0.010849356187763939,-0.010187151299154339,1000,
196+393216,24,pat1-2_q1_gbits2,1-2,2,1,quantizedg,2,0.2850486188865248,-0.06456341907843448,1000,
197+393216,24,pat1-2_q1_gbits24,1-2,2,1,quantizedg,24,0.010848874100210995,-0.010186601043508743,1000,
198+393216,24,pat1-2_q1_gbits3,1-2,2,1,quantizedg,3,0.14026461883203195,-0.0169998463898215,1000,
199+393216,24,pat1-2_q1_gbits4,1-2,2,1,quantizedg,4,0.07008631112291323,-0.011847131509813305,1000,
200+393216,24,pat1-2_q1_gbits52,1-2,2,1,quantizedg,52,0.010848877501410748,-0.01018660322362457,1000,
201+393216,24,pat1-2_q1_gbits6,1-2,2,1,quantizedg,6,0.021096035902636076,-0.011049894366192409,1000,
202+393216,24,pat1-2_q1_gbits8,1-2,2,1,quantizedg,8,0.011943253373206435,-0.010456352092122539,1000,
203+393216,24,pat1-2_q1_single,1-2,2,1,single,64,0.010848877501410873,-0.01018660322362471,1000,
204+393216,24,pat1-2_q1_stale,1-2,2,1,stale,64,0.09057676164114768,-0.08942631958941039,1000,
205+393216,24,pat1-2_q2,1-2,2,2,exact,64,2.3325915027716932e-07,-1.8768761147605335e-07,1000,1.0
206+393216,24,pat1-2_q2_single,1-2,2,2,single,64,2.3325915021999298e-07,-1.8768761140643468e-07,1000,
207+393216,24,pat1-2_q2_stale,1-2,2,2,stale,64,0.0036032717199679866,-0.0035607443091215537,1000,
208+393216,24,pat1-2_q3,1-2,2,3,exact,64,7.361167874105228e-14,1.3923737144427692e-16,1000,1.0
209+393216,24,pat1-2_q3_single,1-2,2,3,single,64,7.364459252706587e-14,6.961868572213838e-17,1000,
210+393216,24,pat1-2_q3_stale,1-2,2,3,stale,64,0.00014387387835912825,-0.00014227754051716342,1000,
211+393216,24,pat1-3_q0,1-3,2,0,exact,64,3.929902012118957,-3.9087825379127272,1000,1.0
212+393216,24,pat1-3_q1,1-3,2,1,exact,64,0.025144743283107493,-0.02462775775994298,1000,1.0
213+393216,24,pat1-3_q2,1-3,2,2,exact,64,1.0897031686530355e-06,-1.008207232245836e-06,1000,1.0
214+393216,24,pat1-3_q3,1-3,2,3,exact,64,7.191310322597904e-14,-1.1835176572763552e-15,1000,1.0
215+393216,24,pat1_q0,1,1,0,exact,64,27.26088657860138,27.259130259724433,1000,1.0
216+393216,24,pat1_q1,1,1,1,exact,64,1.18539309443704,-1.1850887853271908,1000,1.0
217+393216,24,pat1_q1_Hbits12,1,1,1,quantizedH,12,1.1855419187261025,-1.1852362481450442,1000,
218+393216,24,pat1_q1_Hbits16,1,1,1,quantizedH,16,1.185390019085234,-1.1850857126291146,1000,
219+393216,24,pat1_q1_Hbits2,1,1,1,quantizedH,2,1.1405933107136854,1.1372059485897617,1000,
220+393216,24,pat1_q1_Hbits24,1,1,1,quantizedH,24,1.1853930889512956,-1.185088780195468,1000,
221+393216,24,pat1_q1_Hbits3,1,1,1,quantizedH,3,2.59622660213988,-2.5944975244294777,1000,
222+393216,24,pat1_q1_Hbits4,1,1,1,quantizedH,4,0.6107355551832616,-0.6042556721525489,1000,
223+393216,24,pat1_q1_Hbits52,1,1,1,quantizedH,52,1.18539309443704,-1.1850887853271908,1000,
224+393216,24,pat1_q1_Hbits6,1,1,1,quantizedH,6,1.1168986198267827,-1.1112298372923246,1000,
225+393216,24,pat1_q1_Hbits8,1,1,1,quantizedH,8,1.1851843058467242,-1.1844174824673344,1000,
226+393216,24,pat1_q1_bothbits12,1,1,1,quantizedBoth,12,1.1855129744204447,-1.1852038393789903,1000,
227+393216,24,pat1_q1_bothbits16,1,1,1,quantizedBoth,16,1.1853852745973052,-1.1850811740205855,1000,
228+393216,24,pat1_q1_bothbits2,1,1,1,quantizedBoth,2,2.157057674267366,-2.1347466536737034,1000,
229+393216,24,pat1_q1_bothbits24,1,1,1,quantizedBoth,24,1.1853930712088727,-1.1850887624191042,1000,
230+393216,24,pat1_q1_bothbits3,1,1,1,quantizedBoth,3,0.8332332342890458,-0.7462358966217414,1000,
231+393216,24,pat1_q1_bothbits4,1,1,1,quantizedBoth,4,1.418353263772683,-1.293428576685752,1000,
232+393216,24,pat1_q1_bothbits52,1,1,1,quantizedBoth,52,1.1853930944370399,-1.1850887853271905,1000,
233+393216,24,pat1_q1_bothbits6,1,1,1,quantizedBoth,6,1.123258793312136,-1.0999653667060845,1000,
234+393216,24,pat1_q1_bothbits8,1,1,1,quantizedBoth,8,1.1885670321971975,-1.1867955230864966,1000,
235+393216,24,pat1_q1_gbits12,1,1,1,quantizedg,12,1.185364168293707,-1.1850563776150693,1000,
236+393216,24,pat1_q1_gbits16,1,1,1,quantizedg,16,1.1853883498495426,-1.185084246713963,1000,
237+393216,24,pat1_q1_gbits2,1,1,1,quantizedg,2,4.756730271868117,-4.747849551033848,1000,
238+393216,24,pat1_q1_gbits24,1,1,1,quantizedg,24,1.1853930766946332,-1.1850887675508264,1000,
239+393216,24,pat1_q1_gbits3,1,1,1,quantizedg,3,0.672675690610974,0.5759928023587318,1000,
240+393216,24,pat1_q1_gbits4,1,1,1,quantizedg,4,1.9780591455703285,-1.8886240538257422,1000,
241+393216,24,pat1_q1_gbits52,1,1,1,quantizedg,52,1.18539309443704,-1.1850887853271908,1000,
242+393216,24,pat1_q1_gbits6,1,1,1,quantizedg,6,1.1904598965309325,-1.1737833012836951,1000,
243+393216,24,pat1_q1_gbits8,1,1,1,quantizedg,8,1.1887716160175006,-1.1874666976110464,1000,
244+393216,24,pat1_q1_single,1,1,1,single,64,1.1853930944370394,-1.18508878532719,1000,
245+393216,24,pat1_q1_stale,1,1,1,stale,64,2.1811734389616753,2.1806618273552756,1000,
246+393216,24,pat1_q2,1,1,2,exact,64,0.002243075141266478,-0.0022407687140099217,1000,1.0
247+393216,24,pat1_q2_single,1,1,2,single,64,0.002243075141265587,-0.0022407687140090856,1000,
248+393216,24,pat1_q2_stale,1,1,2,stale,64,0.09454716801658596,0.09449581885190024,1000,
249+393216,24,pat1_q3,1,1,3,exact,64,8.056615959372704e-09,-8.023476531210057e-09,1000,1.0
250+393216,24,pat1_q3_single,1,1,3,single,64,8.056615548505713e-09,-8.023476113497943e-09,1000,
251+393216,24,pat1_q3_stale,1,1,3,stale,64,0.00379810147633352,0.0037944826515210503,1000,
252+786432,48,pat1-2-3-4_q0,1-2-3-4,4,0,exact,64,1.5313890370787624,0.03670600496259459,1000,1.0
253+786432,48,pat1-2-3-4_q1,1-2-3-4,4,1,exact,64,0.004511956504111424,-0.002644320606895719,1000,1.0
254+786432,48,pat1-2-3-4_q2,1-2-3-4,4,2,exact,64,6.737707879563051e-08,-2.295468114022171e-08,1000,1.0
255+786432,48,pat1-2-3-4_q3,1-2-3-4,4,3,exact,64,1.0990054327882696e-13,7.482632405255724e-15,1000,1.0
256+786432,48,pat1-2-3_q0,1-2-3,3,0,exact,64,0.6572677044152412,0.051349653946243794,1000,1.0
257+786432,48,pat1-2-3_q1,1-2-3,3,1,exact,64,0.0008190521502529077,-0.00048712137201271927,1000,1.0
258+786432,48,pat1-2-3_q2,1-2-3,3,2,exact,64,2.294364134559273e-09,-7.564363426815919e-10,1000,1.0
259+786432,48,pat1-2-3_q3,1-2-3,3,3,exact,64,1.0972399610735974e-13,6.301164130741662e-15,1000,1.0
260+786432,48,pat1-2-4-8_q0,1-2-4-8,4,0,exact,64,0.5625396329766542,-0.04032064425283584,1000,1.0
261+786432,48,pat1-2-4-8_q1,1-2-4-8,4,1,exact,64,0.0006138888558409122,-0.00035682213864755873,1000,1.0
262+786432,48,pat1-2-4-8_q2,1-2-4-8,4,2,exact,64,1.5202244244390298e-09,-4.248766672100605e-10,1000,1.0
263+786432,48,pat1-2-4-8_q3,1-2-4-8,4,3,exact,64,1.0749269361961354e-13,8.270277921598432e-15,1000,1.0
264+786432,48,pat1-2-4_q0,1-2-4,3,0,exact,64,0.45801712204738726,0.05623087027402669,1000,1.0
265+786432,48,pat1-2-4_q1,1-2-4,3,1,exact,64,0.0004132219541628114,-0.00023655341876729409,1000,1.0
266+786432,48,pat1-2-4_q1_Hbits12,1-2-4,3,1,quantizedH,12,0.0004164912441994055,-0.0002369684421915184,1000,
267+786432,48,pat1-2-4_q1_Hbits16,1-2-4,3,1,quantizedH,16,0.0004139291983452378,-0.00023675825523018608,1000,
268+786432,48,pat1-2-4_q1_Hbits2,1-2-4,3,1,quantizedH,2,0.0010126239936181656,0.00018486121174338723,1000,
269+786432,48,pat1-2-4_q1_Hbits24,1-2-4,3,1,quantizedH,24,0.00041322147562856357,-0.0002365531523929645,1000,
270+786432,48,pat1-2-4_q1_Hbits3,1-2-4,3,1,quantizedH,3,0.0010126239936181656,0.00018486121174338723,1000,
271+786432,48,pat1-2-4_q1_Hbits4,1-2-4,3,1,quantizedH,4,0.0010126239936181656,0.00018486121174338723,1000,
272+786432,48,pat1-2-4_q1_Hbits52,1-2-4,3,1,quantizedH,52,0.0004132219541628114,-0.00023655341876729409,1000,
273+786432,48,pat1-2-4_q1_Hbits6,1-2-4,3,1,quantizedH,6,0.0010181154103176618,0.00017195925746667896,1000,
274+786432,48,pat1-2-4_q1_Hbits8,1-2-4,3,1,quantizedH,8,0.0007692320570511335,-0.00019699168735005275,1000,
275+786432,48,pat1-2-4_q1_bothbits12,1-2-4,3,1,quantizedBoth,12,0.0004181587819740719,-0.00023650609143110825,1000,
276+786432,48,pat1-2-4_q1_bothbits16,1-2-4,3,1,quantizedBoth,16,0.00041371342801174146,-0.00023666284371662513,1000,
277+786432,48,pat1-2-4_q1_bothbits2,1-2-4,3,1,quantizedBoth,2,0.048841631355517785,0.0005229827340559678,1000,
278+786432,48,pat1-2-4_q1_bothbits24,1-2-4,3,1,quantizedBoth,24,0.0004132220297897836,-0.00023655353503066395,1000,
279+786432,48,pat1-2-4_q1_bothbits3,1-2-4,3,1,quantizedBoth,3,0.023819435485863887,0.0006136341969153844,1000,
280+786432,48,pat1-2-4_q1_bothbits4,1-2-4,3,1,quantizedBoth,4,0.012526468977161916,0.0002070504929414913,1000,
281+786432,48,pat1-2-4_q1_bothbits52,1-2-4,3,1,quantizedBoth,52,0.0004132219541628114,-0.00023655341876729409,1000,
282+786432,48,pat1-2-4_q1_bothbits6,1-2-4,3,1,quantizedBoth,6,0.003195328061108964,-1.5213918610698301e-05,1000,
283+786432,48,pat1-2-4_q1_bothbits8,1-2-4,3,1,quantizedBoth,8,0.001126578467953721,-0.0001846208508164093,1000,
284+786432,48,pat1-2-4_q1_gbits12,1-2-4,3,1,quantizedg,12,0.0004144244663849016,-0.00023609092012377817,1000,
285+786432,48,pat1-2-4_q1_gbits16,1-2-4,3,1,quantizedg,16,0.0004130052954246219,-0.0002364580068022138,1000,
286+786432,48,pat1-2-4_q1_gbits2,1-2-4,3,1,quantizedg,2,0.048843227105098225,0.00010439379810680932,1000,
287+786432,48,pat1-2-4_q1_gbits24,1-2-4,3,1,quantizedg,24,0.0004132225083314924,-0.00023655380140824247,1000,
288+786432,48,pat1-2-4_q1_gbits3,1-2-4,3,1,quantizedg,3,0.02380449050784049,0.00019019253604046259,1000,
289+786432,48,pat1-2-4_q1_gbits4,1-2-4,3,1,quantizedg,4,0.012454953685550696,-0.0002132984890702472,1000,
290+786432,48,pat1-2-4_q1_gbits52,1-2-4,3,1,quantizedg,52,0.0004132219541628114,-0.00023655341876729409,1000,
291+786432,48,pat1-2-4_q1_gbits6,1-2-4,3,1,quantizedg,6,0.003129956478607575,-0.00042370871234617425,1000,
292+786432,48,pat1-2-4_q1_gbits8,1-2-4,3,1,quantizedg,8,0.0008730900087727134,-0.00022420326052956913,1000,
293+786432,48,pat1-2-4_q1_single,1-2-4,3,1,single,64,0.0004132219541628114,-0.00023655341876729409,1000,
294+786432,48,pat1-2-4_q1_stale,1-2-4,3,1,stale,64,0.009471325246760514,0.0014216577731255226,1000,
295+786432,48,pat1-2-4_q2,1-2-4,3,2,exact,64,6.086685021192073e-10,-1.925385680903213e-10,1000,1.0
296+786432,48,pat1-2-4_q2_single,1-2-4,3,2,single,64,6.086685021192073e-10,-1.925385680903213e-10,1000,
297+786432,48,pat1-2-4_q2_stale,1-2-4,3,2,stale,64,0.00019391546033417024,2.9992337576640375e-05,1000,
298+786432,48,pat1-2-4_q3,1-2-4,3,3,exact,64,1.1042849128110728e-13,7.876455163427076e-15,1000,1.0
299+786432,48,pat1-2-4_q3_single,1-2-4,3,3,single,64,1.1042849128110728e-13,7.876455163427076e-15,1000,
300+786432,48,pat1-2-4_q3_stale,1-2-4,3,3,stale,64,3.970370453301759e-06,6.302358087135838e-07,1000,
301+786432,48,pat1-2_q0,1-2,2,0,exact,64,0.8847247189682278,-0.8276778089636945,1000,1.0
302+786432,48,pat1-2_q1,1-2,2,1,exact,64,0.001061772620812979,-0.0008825884248293751,1000,1.0
303+786432,48,pat1-2_q1_Hbits12,1-2,2,1,quantizedH,12,0.0010746056228435052,-0.0008864421366251901,1000,
304+786432,48,pat1-2_q1_Hbits16,1-2,2,1,quantizedH,16,0.0010611992769362897,-0.0008822085684174864,1000,
305+786432,48,pat1-2_q1_Hbits2,1-2,2,1,quantizedH,2,0.002234556340236058,0.0007892195224598105,1000,
306+786432,48,pat1-2_q1_Hbits24,1-2,2,1,quantizedH,24,0.0010617703646434353,-0.0008825866212055211,1000,
307+786432,48,pat1-2_q1_Hbits3,1-2,2,1,quantizedH,3,0.002234556340236058,0.0007892195224598105,1000,
308+786432,48,pat1-2_q1_Hbits4,1-2,2,1,quantizedH,4,0.002234556340236058,0.0007892195224598105,1000,
309+786432,48,pat1-2_q1_Hbits52,1-2,2,1,quantizedH,52,0.0010617726208128945,-0.0008825884248292767,1000,
310+786432,48,pat1-2_q1_Hbits6,1-2,2,1,quantizedH,6,0.002234556340236058,0.0007892195224598105,1000,
311+786432,48,pat1-2_q1_Hbits8,1-2,2,1,quantizedH,8,0.001990359547915897,-0.0003911635937278397,1000,
312+786432,48,pat1-2_q1_bothbits12,1-2,2,1,quantizedBoth,12,0.0010763671459987272,-0.0008856807759636977,1000,
313+786432,48,pat1-2_q1_bothbits16,1-2,2,1,quantizedBoth,16,0.0010611145852886651,-0.0008821299698094719,1000,
314+786432,48,pat1-2_q1_bothbits2,1-2,2,1,quantizedBoth,2,0.09536772676529615,-0.01070800822721853,1000,
315+786432,48,pat1-2_q1_bothbits24,1-2,2,1,quantizedBoth,24,0.0010617714222328585,-0.0008825871324297952,1000,
316+786432,48,pat1-2_q1_bothbits3,1-2,2,1,quantizedBoth,3,0.04860667162523445,-0.0007427959881639045,1000,
317+786432,48,pat1-2_q1_bothbits4,1-2,2,1,quantizedBoth,4,0.024248957133891774,0.0011211883770128536,1000,
318+786432,48,pat1-2_q1_bothbits52,1-2,2,1,quantizedBoth,52,0.001061772620812979,-0.0008825884248293751,1000,
319+786432,48,pat1-2_q1_bothbits6,1-2,2,1,quantizedBoth,6,0.006300674592029283,0.0004141598247044639,1000,
320+786432,48,pat1-2_q1_bothbits8,1-2,2,1,quantizedBoth,8,0.00240416369730462,-0.00037522551755773335,1000,
321+786432,48,pat1-2_q1_gbits12,1-2,2,1,quantizedg,12,0.001063863168913454,-0.0008818268011926393,1000,
322+786432,48,pat1-2_q1_gbits16,1-2,2,1,quantizedg,16,0.0010616898946833003,-0.0008825098242106994,1000,
323+786432,48,pat1-2_q1_gbits2,1-2,2,1,quantizedg,2,0.09532392215153007,-0.012358167312931951,1000,
324+786432,48,pat1-2_q1_gbits24,1-2,2,1,quantizedg,24,0.0010617736783632339,-0.0008825889360566072,1000,
325+786432,48,pat1-2_q1_gbits3,1-2,2,1,quantizedg,3,0.04842518581800813,-0.002413715102430527,1000,
326+786432,48,pat1-2_q1_gbits4,1-2,2,1,quantizedg,4,0.024071356768463243,-0.0005519626377464146,1000,
327+786432,48,pat1-2_q1_gbits52,1-2,2,1,quantizedg,52,0.001061772620812979,-0.0008825884248293751,1000,
328+786432,48,pat1-2_q1_gbits6,1-2,2,1,quantizedg,6,0.006118853830385956,-0.0012568517072018281,1000,
329+786432,48,pat1-2_q1_gbits8,1-2,2,1,quantizedg,8,0.0017959843482060224,-0.0008665424565995856,1000,
330+786432,48,pat1-2_q1_single,1-2,2,1,single,64,0.0010617726208130363,-0.0008825884248294736,1000,
331+786432,48,pat1-2_q1_stale,1-2,2,1,stale,64,0.016972531556800993,-0.0159797163899445,1000,
332+786432,48,pat1-2_q2,1-2,2,2,exact,64,2.133394232925136e-09,-1.2711501837389795e-09,1000,1.0
333+786432,48,pat1-2_q2_single,1-2,2,2,single,64,2.1333942130077864e-09,-1.27115008528329e-09,1000,
334+786432,48,pat1-2_q2_stale,1-2,2,2,stale,64,0.0003446194228576056,-0.00032497866972252523,1000,
335+786432,48,pat1-2_q3,1-2,2,3,exact,64,1.0690492685164926e-13,5.218151545770439e-15,1000,1.0
336+786432,48,pat1-2_q3_single,1-2,2,3,single,64,1.0685958011499542e-13,5.1196958562276e-15,1000,
337+786432,48,pat1-2_q3_stale,1-2,2,3,stale,64,7.006064854080526e-06,-6.616146066708182e-06,1000,
338+786432,48,pat1-3_q0,1-3,2,0,exact,64,1.296217033355432,-1.2671915404186673,1000,1.0
339+786432,48,pat1-3_q1,1-3,2,1,exact,64,0.0020458190356249847,-0.0018945564842300596,1000,1.0
340+786432,48,pat1-3_q2,1-3,2,2,exact,64,6.034399971284161e-09,-4.719451128794436e-09,1000,1.0
341+786432,48,pat1-3_q3,1-3,2,3,exact,64,1.0963561591089451e-13,1.1027037228797908e-14,1000,1.0
342+786432,48,pat1_q0,1,1,0,exact,64,18.868778074260554,18.867498044474743,1000,1.0
343+786432,48,pat1_q1,1,1,1,exact,64,0.4015638968530596,-0.4014557369106377,1000,1.0
344+786432,48,pat1_q1_Hbits12,1,1,1,quantizedH,12,0.40154384767579154,-0.40143280730319564,1000,
345+786432,48,pat1_q1_Hbits16,1,1,1,quantizedH,16,0.401560559488193,-0.40145232009336956,1000,
346+786432,48,pat1_q1_Hbits2,1,1,1,quantizedH,2,0.397073952406543,0.3946485955504654,1000,
347+786432,48,pat1_q1_Hbits24,1,1,1,quantizedH,24,0.4015639083815828,-0.40145574863171846,1000,
348+786432,48,pat1_q1_Hbits3,1,1,1,quantizedH,3,0.397073952406543,0.3946485955504654,1000,
349+786432,48,pat1_q1_Hbits4,1,1,1,quantizedH,4,0.8380436716771879,-0.8368747010444832,1000,
350+786432,48,pat1_q1_Hbits52,1,1,1,quantizedH,52,0.40156389685305893,-0.401455736910637,1000,
351+786432,48,pat1_q1_Hbits6,1,1,1,quantizedH,6,0.47394415805223933,-0.46520060481281533,1000,
352+786432,48,pat1_q1_Hbits8,1,1,1,quantizedH,8,0.4023425395711377,-0.40158871281994907,1000,
353+786432,48,pat1_q1_bothbits12,1,1,1,quantizedBoth,12,0.4014508897008998,-0.40133416155905005,1000,
354+786432,48,pat1_q1_bothbits16,1,1,1,quantizedBoth,16,0.40156866283048254,-0.4014604030424933,1000,
355+786432,48,pat1_q1_bothbits2,1,1,1,quantizedBoth,2,1.929668526461305,-1.9171116463517859,1000,
356+786432,48,pat1_q1_bothbits24,1,1,1,quantizedBoth,24,0.40156385610421275,-0.40145569629896716,1000,
357+786432,48,pat1_q1_bothbits3,1,1,1,quantizedBoth,3,1.5560619231640465,1.526205359095142,1000,
358+786432,48,pat1_q1_bothbits4,1,1,1,quantizedBoth,4,1.3868834574507904,-1.324057449840225,1000,
359+786432,48,pat1_q1_bothbits52,1,1,1,quantizedBoth,52,0.4015638968530594,-0.40145573691063746,1000,
360+786432,48,pat1_q1_bothbits6,1,1,1,quantizedBoth,6,0.4837147965393759,-0.45647347511226877,1000,
361+786432,48,pat1_q1_bothbits8,1,1,1,quantizedBoth,8,0.4035724964206338,-0.4014884719445613,1000,
362+786432,48,pat1_q1_gbits12,1,1,1,quantizedg,12,0.4014707672986577,-0.40135708757267713,1000,
363+786432,48,pat1_q1_gbits16,1,1,1,quantizedg,16,0.40157199963264817,-0.4014638198496341,1000,
364+786432,48,pat1_q1_gbits2,1,1,1,quantizedg,2,2.820524764962026,-2.8127691412286118,1000,
365+786432,48,pat1_q1_gbits24,1,1,1,quantizedg,24,0.40156384457571254,-0.40145568457788655,1000,
366+786432,48,pat1_q1_gbits3,1,1,1,quantizedg,3,0.8372806982707423,0.7788941808994515,1000,
367+786432,48,pat1_q1_gbits4,1,1,1,quantizedg,4,0.9657946478580162,-0.8779052024247551,1000,
368+786432,48,pat1_q1_gbits52,1,1,1,quantizedg,52,0.4015638968530595,-0.4014557369106376,1000,
369+786432,48,pat1_q1_gbits6,1,1,1,quantizedg,6,0.4146255655979421,-0.3927648263383849,1000,
370+786432,48,pat1_q1_gbits8,1,1,1,quantizedg,8,0.4027866605873682,-0.40135532406792096,1000,
371+786432,48,pat1_q1_single,1,1,1,single,64,0.4015638968530598,-0.4014557369106379,1000,
372+786432,48,pat1_q1_stale,1,1,1,stale,64,0.7703185454020058,0.7701213594019068,1000,
373+786432,48,pat1_q2,1,1,2,exact,64,0.0001820236580114578,-0.0001818273384480346,1000,1.0
374+786432,48,pat1_q2_single,1,1,2,single,64,0.00018202365801071922,-0.0001818273384472466,1000,
375+786432,48,pat1_q2_stale,1,1,2,stale,64,0.016383210808024264,0.01637367779242386,1000,
376+786432,48,pat1_q3,1,1,3,exact,64,3.751853378425974e-11,-3.7355073169448336e-11,1000,1.0
377+786432,48,pat1_q3_single,1,1,3,single,64,3.7519027260548054e-11,-3.735546699220651e-11,1000,
378+786432,48,pat1_q3_stale,1,1,3,stale,64,0.000334882920486108,0.00033453988160991573,1000,
379+1572864,96,pat1-2-3-4_q0,1-2-3-4,4,0,exact,64,1.0033885259698323,0.05157913141748878,200,1.0
380+1572864,96,pat1-2-3-4_q1,1-2-3-4,4,1,exact,64,0.0014202144516986125,-0.000802756544389786,200,1.0
381+1572864,96,pat1-2-3-4_q2,1-2-3-4,4,2,exact,64,5.561541569528131e-09,-1.6082187914688319e-09,200,1.0
382+1572864,96,pat1-2-3-4_q3,1-2-3-4,4,3,exact,64,1.4335364781935019e-13,-2.7847474288855425e-15,200,1.0
383+1572864,96,pat1-2-3_q0,1-2-3,3,0,exact,64,0.4458187645235926,0.06561168563196455,200,1.0
384+1572864,96,pat1-2-3_q1,1-2-3,3,1,exact,64,0.0002702812332041187,-0.000158477380747302,200,1.0
385+1572864,96,pat1-2-3_q2,1-2-3,3,2,exact,64,1.7467885132797996e-10,-5.824577722256998e-11,200,1.0
386+1572864,96,pat1-2-3_q3,1-2-3,3,3,exact,64,1.537927034686241e-13,2.7847474288855413e-15,200,1.0
387+1572864,96,pat1-2-4-8_q0,1-2-4-8,4,0,exact,64,0.38702806624821595,0.0724006263517435,200,1.0
388+1572864,96,pat1-2-4-8_q1,1-2-4-8,4,1,exact,64,0.00019783861243691123,-0.00011943761958857218,200,1.0
389+1572864,96,pat1-2-4-8_q2,1-2-4-8,4,2,exact,64,8.243874438987699e-11,-3.119961400637638e-11,200,1.0
390+1572864,96,pat1-2-4-8_q3,1-2-4-8,4,3,exact,64,1.5252689837551246e-13,9.746616001099394e-15,200,1.0
391+1572864,96,pat1-2-4_q0,1-2-4,3,0,exact,64,0.31558872194341686,0.07028920370293829,200,1.0
392+1572864,96,pat1-2-4_q1,1-2-4,3,1,exact,64,0.00013019947341526842,-7.941538160455175e-05,200,1.0
393+1572864,96,pat1-2-4_q1_Hbits12,1-2-4,3,1,quantizedH,12,0.0001414099966356698,-8.19537632833217e-05,200,
394+1572864,96,pat1-2-4_q1_Hbits16,1-2-4,3,1,quantizedH,16,0.00012992318308216742,-7.940038452689105e-05,200,
395+1572864,96,pat1-2-4_q1_Hbits2,1-2-4,3,1,quantizedH,2,0.00047373810965483217,6.918890390049704e-05,200,
396+1572864,96,pat1-2-4_q1_Hbits24,1-2-4,3,1,quantizedH,24,0.00013020185962545546,-7.94158143480375e-05,200,
397+1572864,96,pat1-2-4_q1_Hbits3,1-2-4,3,1,quantizedH,3,0.00047373810965483217,6.918890390049704e-05,200,
398+1572864,96,pat1-2-4_q1_Hbits4,1-2-4,3,1,quantizedH,4,0.00047373810965483217,6.918890390049704e-05,200,
399+1572864,96,pat1-2-4_q1_Hbits52,1-2-4,3,1,quantizedH,52,0.00013019947341526842,-7.941538160455175e-05,200,
400+1572864,96,pat1-2-4_q1_Hbits6,1-2-4,3,1,quantizedH,6,0.00047373810965483217,6.918890390049704e-05,200,
401+1572864,96,pat1-2-4_q1_Hbits8,1-2-4,3,1,quantizedH,8,0.00043279428797429975,-3.8397998901916246e-05,200,
402+1572864,96,pat1-2-4_q1_bothbits12,1-2-4,3,1,quantizedBoth,12,0.00014377297203402667,-8.263002829692466e-05,200,
403+1572864,96,pat1-2-4_q1_bothbits16,1-2-4,3,1,quantizedBoth,16,0.00012952417413678602,-7.93053343803298e-05,200,
404+1572864,96,pat1-2-4_q1_bothbits2,1-2-4,3,1,quantizedBoth,2,0.03340790435060609,0.003729582518423993,200,
405+1572864,96,pat1-2-4_q1_bothbits24,1-2-4,3,1,quantizedBoth,24,0.00013020160450507405,-7.941529398734803e-05,200,
406+1572864,96,pat1-2-4_q1_bothbits3,1-2-4,3,1,quantizedBoth,3,0.017193550142416577,-0.0004206759813111777,200,
407+1572864,96,pat1-2-4_q1_bothbits4,1-2-4,3,1,quantizedBoth,4,0.008228711914540249,9.960138104755508e-05,200,
408+1572864,96,pat1-2-4_q1_bothbits52,1-2-4,3,1,quantizedBoth,52,0.00013019947341526842,-7.941538160455175e-05,200,
409+1572864,96,pat1-2-4_q1_bothbits6,1-2-4,3,1,quantizedBoth,6,0.002316001949632785,3.643552311750455e-05,200,
410+1572864,96,pat1-2-4_q1_bothbits8,1-2-4,3,1,quantizedBoth,8,0.0007255545715923543,4.021418842991267e-05,200,
411+1572864,96,pat1-2-4_q1_gbits12,1-2-4,3,1,quantizedg,12,0.00013250501497750688,-8.009142274325908e-05,200,
412+1572864,96,pat1-2-4_q1_gbits16,1-2-4,3,1,quantizedg,16,0.00012980027227848005,-7.93203328447984e-05,200,
413+1572864,96,pat1-2-4_q1_gbits2,1-2-4,3,1,quantizedg,2,0.033397209236530494,0.0035798005156970013,200,
414+1572864,96,pat1-2-4_q1_gbits24,1-2-4,3,1,quantizedg,24,0.00013019921835916553,-7.941486125917752e-05,200,
415+1572864,96,pat1-2-4_q1_gbits3,1-2-4,3,1,quantizedg,3,0.017163773301913317,-0.0005664988972504842,200,
416+1572864,96,pat1-2-4_q1_gbits4,1-2-4,3,1,quantizedg,4,0.008185233884679049,-4.895176367652143e-05,200,
417+1572864,96,pat1-2-4_q1_gbits52,1-2-4,3,1,quantizedg,52,0.00013019947341526842,-7.941538160455175e-05,200,
418+1572864,96,pat1-2-4_q1_gbits6,1-2-4,3,1,quantizedg,6,0.0022733568235961377,-0.0001124833019683328,200,
419+1572864,96,pat1-2-4_q1_gbits8,1-2-4,3,1,quantizedg,8,0.0005469932582039279,-8.527460364573427e-07,200,
420+1572864,96,pat1-2-4_q1_single,1-2-4,3,1,single,64,0.00013019947341526842,-7.941538160455175e-05,200,
421+1572864,96,pat1-2-4_q1_stale,1-2-4,3,1,stale,64,0.0033177727033229187,0.0008193548488570263,200,
422+1572864,96,pat1-2-4_q2,1-2-4,3,2,exact,64,3.4381194816229206e-11,-1.3522733514668189e-11,200,1.0
423+1572864,96,pat1-2-4_q2_single,1-2-4,3,2,single,64,3.4381194816229206e-11,-1.3522733514668189e-11,200,
424+1572864,96,pat1-2-4_q2_stale,1-2-4,3,2,stale,64,3.432233628387408e-05,8.624766025648093e-06,200,
425+1572864,96,pat1-2-4_q3,1-2-4,3,3,exact,64,1.5060824466094994e-13,5.5694948577710826e-15,200,1.0
426+1572864,96,pat1-2-4_q3_single,1-2-4,3,3,single,64,1.5060824466094994e-13,5.5694948577710826e-15,200,
427+1572864,96,pat1-2-4_q3_stale,1-2-4,3,3,stale,64,3.5508870973618843e-07,9.067003055531822e-08,200,
428+1572864,96,pat1-2_q0,1-2,2,0,exact,64,0.32275202501920325,-0.24400887177078898,200,1.0
429+1572864,96,pat1-2_q1,1-2,2,1,exact,64,0.00012872529978105965,-8.30588283739578e-05,200,1.0
430+1572864,96,pat1-2_q1_Hbits12,1-2,2,1,quantizedH,12,0.00012901740662731904,-8.040002648938526e-05,200,
431+1572864,96,pat1-2_q1_Hbits16,1-2,2,1,quantizedH,16,0.0001287760636558924,-8.310663756109046e-05,200,
432+1572864,96,pat1-2_q1_Hbits2,1-2,2,1,quantizedH,2,0.0005198462334860989,6.97062799378879e-05,200,
433+1572864,96,pat1-2_q1_Hbits24,1-2,2,1,quantizedH,24,0.0001287270006927484,-8.305922051025495e-05,200,
434+1572864,96,pat1-2_q1_Hbits3,1-2,2,1,quantizedH,3,0.0005198462334860989,6.97062799378879e-05,200,
435+1572864,96,pat1-2_q1_Hbits4,1-2,2,1,quantizedH,4,0.0005198462334860989,6.97062799378879e-05,200,
436+1572864,96,pat1-2_q1_Hbits52,1-2,2,1,quantizedH,52,0.00012872529978105965,-8.30588283739578e-05,200,
437+1572864,96,pat1-2_q1_Hbits6,1-2,2,1,quantizedH,6,0.0005198462334860989,6.97062799378879e-05,200,
438+1572864,96,pat1-2_q1_Hbits8,1-2,2,1,quantizedH,8,0.0004800970126231751,9.020467736765304e-05,200,
439+1572864,96,pat1-2_q1_bothbits12,1-2,2,1,quantizedBoth,12,0.0001384610679661378,-8.109653317624083e-05,200,
440+1572864,96,pat1-2_q1_bothbits16,1-2,2,1,quantizedBoth,16,0.00012868215158162426,-8.305919158647522e-05,200,
441+1572864,96,pat1-2_q1_bothbits2,1-2,2,1,quantizedBoth,2,0.03369201416428462,-0.004311073356307374,200,
442+1572864,96,pat1-2_q1_bothbits24,1-2,2,1,quantizedBoth,24,0.00012872629839211648,-8.305921934205514e-05,200,
443+1572864,96,pat1-2_q1_bothbits3,1-2,2,1,quantizedBoth,3,0.017441597274721006,-0.0004896323921570679,200,
444+1572864,96,pat1-2_q1_bothbits4,1-2,2,1,quantizedBoth,4,0.008153799329594176,-8.452059177878709e-05,200,
445+1572864,96,pat1-2_q1_bothbits52,1-2,2,1,quantizedBoth,52,0.00012872529978105965,-8.30588283739578e-05,200,
446+1572864,96,pat1-2_q1_bothbits6,1-2,2,1,quantizedBoth,6,0.0022192205440188296,-7.368148006298288e-05,200,
447+1572864,96,pat1-2_q1_bothbits8,1-2,2,1,quantizedBoth,8,0.0006710706433711802,0.00010273147277383873,200,
448+1572864,96,pat1-2_q1_gbits12,1-2,2,1,quantizedg,12,0.00013739910288517034,-8.375557504477455e-05,200,
449+1572864,96,pat1-2_q1_gbits16,1-2,2,1,quantizedg,16,0.00012862667677442276,-8.301138323406787e-05,200,
450+1572864,96,pat1-2_q1_gbits2,1-2,2,1,quantizedg,2,0.033776779160772444,-0.0044556509852573395,200,
451+1572864,96,pat1-2_q1_gbits24,1-2,2,1,quantizedg,24,0.00012872459760442708,-8.305882720854274e-05,200,
452+1572864,96,pat1-2_q1_gbits3,1-2,2,1,quantizedg,3,0.01742299470910743,-0.0006421244686534333,200,
453+1572864,96,pat1-2_q1_gbits4,1-2,2,1,quantizedg,4,0.008200905281267531,-0.00023574075403911244,200,
454+1572864,96,pat1-2_q1_gbits52,1-2,2,1,quantizedg,52,0.00012872529978105965,-8.30588283739578e-05,200,
455+1572864,96,pat1-2_q1_gbits6,1-2,2,1,quantizedg,6,0.0021942950436127915,-0.0002263176447958203,200,
456+1572864,96,pat1-2_q1_gbits8,1-2,2,1,quantizedg,8,0.0005352895046896114,-7.04846588556651e-05,200,
457+1572864,96,pat1-2_q1_single,1-2,2,1,single,64,0.00012872529978105965,-8.30588283739578e-05,200,
458+1572864,96,pat1-2_q1_stale,1-2,2,1,stale,64,0.0031894157505053854,-0.0024220620218652483,200,
459+1572864,96,pat1-2_q2,1-2,2,2,exact,64,3.321136472128824e-11,-1.3220588418634107e-11,200,1.0
460+1572864,96,pat1-2_q2_single,1-2,2,2,single,64,3.321136472128824e-11,-1.3220588418634107e-11,200,
461+1572864,96,pat1-2_q2_stale,1-2,2,2,stale,64,3.26839528598533e-05,-2.4853796671434528e-05,200,
462+1572864,96,pat1-2_q3,1-2,2,3,exact,64,1.569125486394371e-13,4.177121143328311e-15,200,1.0
463+1572864,96,pat1-2_q3_single,1-2,2,3,single,64,1.569125486394371e-13,4.177121143328311e-15,200,
464+1572864,96,pat1-2_q3_stale,1-2,2,3,stale,64,3.351238929895299e-07,-2.5514060274291926e-07,200,
465+1572864,96,pat1-3_q0,1-3,2,0,exact,64,0.4370905705959195,-0.39881915047203975,200,1.0
466+1572864,96,pat1-3_q1,1-3,2,1,exact,64,0.00019399329290125576,-0.00015233086292993646,200,1.0
467+1572864,96,pat1-3_q2,1-3,2,2,exact,64,5.275146970009042e-11,-3.0002172611955604e-11,200,1.0
468+1572864,96,pat1-3_q3,1-3,2,3,exact,64,1.6415843990891575e-13,-5.5694948577710826e-15,200,1.0
469+1572864,96,pat1_q0,1,1,0,exact,64,13.222904223779382,13.222110671551855,200,1.0
470+1572864,96,pat1_q1,1,1,1,exact,64,0.13944458967580223,-0.13941146509877655,200,1.0
471+1572864,96,pat1_q1_Hbits12,1,1,1,quantizedH,12,0.13946805961741204,-0.13943022049734005,200,
472+1572864,96,pat1_q1_Hbits16,1,1,1,quantizedH,16,0.13943646515412692,-0.13940334920566852,200,
473+1572864,96,pat1_q1_Hbits2,1,1,1,quantizedH,2,0.14013371937098454,0.13849855287179613,200,
474+1572864,96,pat1_q1_Hbits24,1,1,1,quantizedH,24,0.1394445914451559,-0.13941146650984915,200,
475+1572864,96,pat1_q1_Hbits3,1,1,1,quantizedH,3,0.14013371937098454,0.13849855287179613,200,
476+1572864,96,pat1_q1_Hbits4,1,1,1,quantizedH,4,0.14013371937098454,0.13849855287179613,200,
477+1572864,96,pat1_q1_Hbits52,1,1,1,quantizedH,52,0.13944458967580223,-0.13941146509877655,200,
478+1572864,96,pat1_q1_Hbits6,1,1,1,quantizedH,6,0.091415588806578,-0.08097239749050143,200,
479+1572864,96,pat1_q1_Hbits8,1,1,1,quantizedH,8,0.1404244271775348,-0.1395282715617766,200,
480+1572864,96,pat1_q1_bothbits12,1,1,1,quantizedBoth,12,0.13939161296729313,-0.13934489857049592,200,
481+1572864,96,pat1_q1_bothbits16,1,1,1,quantizedBoth,16,0.13944254553680377,-0.13940951076191915,200,
482+1572864,96,pat1_q1_bothbits2,1,1,1,quantizedBoth,2,1.4819252667794753,-1.4748277851472131,200,
483+1572864,96,pat1_q1_bothbits24,1,1,1,quantizedBoth,24,0.13944462222898005,-0.13941149708954356,200,
484+1572864,96,pat1_q1_bothbits3,1,1,1,quantizedBoth,3,0.9853686782823758,0.9746619576359649,200,
485+1572864,96,pat1_q1_bothbits4,1,1,1,quantizedBoth,4,0.31002387061174275,-0.16435077275821286,200,
486+1572864,96,pat1_q1_bothbits52,1,1,1,quantizedBoth,52,0.13944458967580223,-0.13941146509877655,200,
487+1572864,96,pat1_q1_bothbits6,1,1,1,quantizedBoth,6,0.12938100209490971,-0.079157945938691,200,
488+1572864,96,pat1_q1_bothbits8,1,1,1,quantizedBoth,8,0.14279283384708125,-0.1400331788227848,200,
489+1572864,96,pat1_q1_gbits12,1,1,1,quantizedg,12,0.13936843816462097,-0.1393261464283868,200,
490+1572864,96,pat1_q1_gbits16,1,1,1,quantizedg,16,0.13945066680138404,-0.13941762662559376,200,
491+1572864,96,pat1_q1_gbits2,1,1,1,quantizedg,2,1.7924752607323675,-1.786974179064042,200,
492+1572864,96,pat1_q1_gbits24,1,1,1,quantizedg,24,0.13944462045969147,-0.13941149567846192,200,
493+1572864,96,pat1_q1_gbits3,1,1,1,quantizedg,3,0.7282798807180506,0.7145399627052711,200,
494+1572864,96,pat1_q1_gbits4,1,1,1,quantizedg,4,0.5223760212256646,-0.44868523960100426,200,
495+1572864,96,pat1_q1_gbits52,1,1,1,quantizedg,52,0.13944458967580223,-0.13941146509877655,200,
496+1572864,96,pat1_q1_gbits6,1,1,1,quantizedg,6,0.16472851277777803,-0.13756908684207095,200,
497+1572864,96,pat1_q1_gbits8,1,1,1,quantizedg,8,0.1417224457909359,-0.13991522670068238,200,
498+1572864,96,pat1_q1_single,1,1,1,single,64,0.13944458967580226,-0.13941146509877655,200,
499+1572864,96,pat1_q1_stale,1,1,1,stale,64,0.27311641471218,0.27305354912416463,200,
500+1572864,96,pat1_q2,1,1,2,exact,64,1.5518767391080547e-05,-1.5504140223044486e-05,200,1.0
501+1572864,96,pat1_q2_single,1,1,2,single,64,1.551876739114228e-05,-1.5504140223044486e-05,200,
502+1572864,96,pat1_q2_stale,1,1,2,stale,64,0.002879989299924687,0.0028785043284346133,200,
503+1572864,96,pat1_q3,1,1,3,exact,64,2.417683653918259e-13,-1.8866663830699544e-13,200,1.0
504+1572864,96,pat1_q3_single,1,1,3,single,64,2.421689760483422e-13,-1.9005901202143819e-13,200,
505+1572864,96,pat1_q3_stale,1,1,3,stale,64,2.9761423694967157e-05,2.973445994530665e-05,200,
added results/processed/ablation_logistic_summary.csv +128 −0
Large diff (128 lines) is not loaded by default to keep the page fast.Load diff
added results/processed/boundary_summary_with_prediction.csv +76 −0
Large diff (76 lines) is not loaded by default to keep the page fast.Load diff
added results/processed/skew_summary.csv +37 −0
Large diff (37 lines) is not loaded by default to keep the page fast.Load diff
modified results/processed/summary.csv +726 −0
Large diff (739 lines) is not loaded by default to keep the page fast.Load diff
added results/raw/ablation_exponential_N1572864_m96.csv +17516 −0
Large diff (17,516 lines) is not loaded by default to keep the page fast.Load diff

Diff truncated — file too large.

Menu

Theme
↑↓ navigate ↵ open esc close / or ⌘K anywhere