matCercola18 commited on
Commit
f7150e6
·
verified ·
1 Parent(s): 95ebf14

X28-M: spline with identity fallback, fresh test plants T2

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. results/icml_r3/X28/PREREG.md +30 -0
  2. results/icml_r3/X28/mimo/fb_fresh_t2_m2.npz +3 -0
  3. results/icml_r3/X28/mimo/fb_fresh_t2_m2.txt +10 -0
  4. results/icml_r3/X28/mimo/fb_fresh_t2_m3.npz +3 -0
  5. results/icml_r3/X28/mimo/fb_fresh_t2_m3.txt +10 -0
  6. results/icml_r3/X28/mimo/fb_heldout_m2.txt +30 -0
  7. results/icml_r3/X28/mimo/fb_heldout_m3.txt +29 -0
  8. results/icml_r3/X28/mimo/fb_paired_t2_m2.txt +68 -0
  9. results/icml_r3/X28/mimo/fb_paired_t2_m3.txt +68 -0
  10. results/icml_r3/X28/mimo/fb_val_choice.txt +38 -0
  11. results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m2a_s0.npz +3 -0
  12. results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m2a_s1.npz +3 -0
  13. results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m2a_s2.npz +3 -0
  14. results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m3a_s0.npz +3 -0
  15. results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m3a_s1.npz +3 -0
  16. results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m3a_s2.npz +3 -0
  17. results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m2a_s0.npz +3 -0
  18. results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m2a_s1.npz +3 -0
  19. results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m2a_s2.npz +3 -0
  20. results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m3a_s0.npz +3 -0
  21. results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m3a_s1.npz +3 -0
  22. results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m3a_s2.npz +3 -0
  23. results/icml_r3/X28/mimo/gen_fb/t2_c_x28m2c_s0.npz +3 -0
  24. results/icml_r3/X28/mimo/gen_fb/t2_c_x28m2c_s1.npz +3 -0
  25. results/icml_r3/X28/mimo/gen_fb/t2_c_x28m2c_s2.npz +3 -0
  26. results/icml_r3/X28/mimo/gen_fb/t2_c_x28m3c_s0.npz +3 -0
  27. results/icml_r3/X28/mimo/gen_fb/t2_c_x28m3c_s1.npz +3 -0
  28. results/icml_r3/X28/mimo/gen_fb/t2_c_x28m3c_s2.npz +3 -0
  29. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2a_s0.npz +3 -0
  30. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2a_s1.npz +3 -0
  31. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2a_s2.npz +3 -0
  32. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2b_s0.npz +3 -0
  33. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2b_s1.npz +3 -0
  34. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2b_s2.npz +3 -0
  35. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3a_s0.npz +3 -0
  36. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3a_s1.npz +3 -0
  37. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3a_s2.npz +3 -0
  38. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3b_s0.npz +3 -0
  39. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3b_s1.npz +3 -0
  40. results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3b_s2.npz +3 -0
  41. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2a_s0.npz +3 -0
  42. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2a_s1.npz +3 -0
  43. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2a_s2.npz +3 -0
  44. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2b_s0.npz +3 -0
  45. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2b_s1.npz +3 -0
  46. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2b_s2.npz +3 -0
  47. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3a_s0.npz +3 -0
  48. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3a_s1.npz +3 -0
  49. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3a_s2.npz +3 -0
  50. results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3b_s0.npz +3 -0
results/icml_r3/X28/PREREG.md CHANGED
@@ -249,3 +249,33 @@ Secondaries, reported with the primary: (i) the either-or certificate without re
249
  Outputs: `mimo/gen_rep/`, `mimo/gen_rep_paired_m{2,3}.txt`, `mimo/gen_rep_fresh_m{2,3}.txt`, `mimo/rep_selfcheck.txt`, `mimo/rep_val_choice.txt`.
250
 
251
  Compute: the CPU node `move-hpc-01-cpu` (back up), in one process, because the /home NFS is slow to import from.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
249
  Outputs: `mimo/gen_rep/`, `mimo/gen_rep_paired_m{2,3}.txt`, `mimo/gen_rep_fresh_m{2,3}.txt`, `mimo/rep_selfcheck.txt`, `mimo/rep_val_choice.txt`.
250
 
251
  Compute: the CPU node `move-hpc-01-cpu` (back up), in one process, because the /home NFS is slow to import from.
252
+
253
+ ## Addendum 2026-11-15: MIMO spline with identity fallback, on fresh test plants
254
+
255
+ Written after the S4-M with repair result (`gen_rep_paired_m*.txt`) and its diagnostic (`rep_tails.txt`), and before any code, sample or number of this test exists. The S4-M verdict stays the registered primary result of X28, and S4-M with repair stays as reported.
256
+
257
+ **Why** (Matteo, 2026-11-15: "Can you try to improve the spline on decision 1"; decision 1 was a fix for the plants where the spline does worse). With repair, the spline generator wins at m = 3 and is 1.9 to 2.5 points cheaper at m = 2, but on 12 to 19 held-out plants per seed it is more than 10 points worse than identity (`rep_tails.txt`). On those plants a median of 2 to 6 of its 16 samples certify as sampled (10 to 16 on the other plants), and its best controller has 26% to 41% regret against 7% to 21% for identity. How many samples certify is known at run time, so a few spline samples can decide where to use the identity generator instead.
258
+ The variant was designed from the held-out diagnostic, so its verdict comes from new plants (T2) that no model, rule or diagnostic has seen.
259
+
260
+ **Arm (c): spline generator with identity fallback.** Per plant and training seed s, with a budget of b = 16 generated samples:
261
+ 1. Probe: the first q samples of generator (b)_s (the draws of arm (b)), each tested with the circle LMI as sampled.
262
+ 2. If at most c of the q probe samples are certified as sampled, the other 16 - q samples come from generator (a)_s, drawn with sampler seed = model seed + 1000, independent of arm (a)'s own draws. Otherwise they are samples q + 1 to 16 of (b)_s.
263
+ 3. Spline samples get repair R of the 2026-11-14 addendum as chosen there (circle LMI, center r_i(1)). Identity samples get the Lyapunov test of arm (a).
264
+ 4. As in every arm, the certified candidate with the lowest J is used. There is no other cost-based choice, and every plant uses 16 generated samples.
265
+ With q >= 4 the b = 1 and b = 4 results of (c) equal those of (b) with repair, because the probe comes first. b = 16 is the primary.
266
+ Grid: q in {4, 8}, c in {-1, 0, 1, ..., q}. c = -1 never falls back (arm (b) with repair); c = q always does.
267
+
268
+ **Choice of (q, c) on validation, per m.** The 64 validation plants [0, 64), their J*_val and their stored samples (`gen_rep/val_cand_*.npz`, addendum 2026-11-14), plus fallback draws (seed + 1000) on the same plants. Pooled over the 3 seed pairs (192 plant-seed pairs per m), d = regret (a) - regret (c) at b = 16, uncertified = 1e3. Eligible: plant-level certified rate of (c) at most 2 points below that of (a). Choice: the eligible (q, c) with the largest standardized Wilcoxon signed-rank statistic of d (zeros dropped, average ranks, tie-corrected variance), because the Wilcoxon p is what stopped m = 2. Ties: smaller c, then smaller q. If none is eligible: c = -1. Saved to `mimo/fb_val_choice.txt` before any T2 regret is computed.
269
+
270
+ **Fresh test plants T2.** Built like the held-out sets (`build_sofm`, kind heldout: uniform in the family box, 48 oracle starts, kept iff the oracle certifies at least one gain), builder seeds 1115 and 1116, 100 plants each: 200 plants per m, never in a bank, validation or test set. K0 = the best oracle gain.
271
+ J*_T2: the S2-M label optimizer on the T2 plants (`id` at c = 1, `so3` at c = 1, 2, 5, K lr 0.02, knot lr 0.02, 400 steps, snapshot every 25, cheapest certified snapshot, `X28M_SEG=100`); J* = min over these four. The held-out J* also includes the `sp` and `so` classes; T2 uses the classes the generators learned.
272
+ Generators on T2: the 12 S3-M models, 16 samples per plant, DDPM-500, clip 1, sampler seed = model seed, on the GPU (as the held-out run). Fallback draws: generators (a)_s, 12 samples per plant, seed = model seed + 1000. Certificates and cost J as in S4-M.
273
+
274
+ **Tests on T2, per m.** The S4-M rule: b = 16; "better" iff median paired regret difference >= 1 point, Holm (3 seed pairs) Wilcoxon p < 0.05, and certified rate not lower by more than 2 points; verdict iff on >= 2 of 3 pairs; b = 1 and 4 reported.
275
+ - Primary: (c) at the chosen (q, c) against (a), seed j against seed j.
276
+ - Replications (secondary): (b) with repair R against (a), the 2026-11-14 test; (b) as sampled against (a), the 2026-11-13 test.
277
+ - Reported: the fallback rate per seed; the `rep_tails.txt` table for (c); fresh directions, each arm's best certified b = 16 controller per plant on 64 random unit directions (seed (7979, plant index)), against (a) seed 0.
278
+ - Also reported, labelled as the plants that shaped the variant: (c) on the 120 held-out plants, with the stored samples plus fallback draws (seed + 1000) there.
279
+ Outputs: `mimo/t2/` (plants), `mimo/t2_m{2,3}/` (labels), `mimo/gen_fb/` (samples, evaluations), `mimo/fb_val_choice.txt`, `mimo/fb_paired_t2_m{2,3}.txt`, `mimo/fb_fresh_t2_m{2,3}.txt`, `mimo/fb_heldout_m{2,3}.txt`.
280
+
281
+ Compute: the CPU node is down, and node 01 stays free for Matteo's robot session on Monday. One GPU node lent by GPL_move, one Slurm job: the GPU for the label optimizer, the samplers and the cost rollouts, its CPUs for the oracle and the LMI pool.
results/icml_r3/X28/mimo/fb_fresh_t2_m2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8dc6fff2d88502416950b63cc4a73dd86af6dc054dda9b161d80bfb4997b787e
3
+ size 1763653
results/icml_r3/X28/mimo/fb_fresh_t2_m2.txt ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback, fresh directions on T2, m = 2: best certified b=16 sample per plant on 64 unit directions (seed (7979, plant index)), vs x28m2a_s0; _rep = (b) with repair, c = fallback arm
2
+ x28m2a_s0: median over plants of per-plant median (J_a0 - J) / J +0.00% [+0.00, +0.00], > 0 on 0 of 199; non-convergent 5 of 12736
3
+ x28m2a_s1: median over plants of per-plant median (J_a0 - J) / J -0.04% [-0.15, +0.08], > 0 on 95 of 199; non-convergent 0 of 12800
4
+ x28m2a_s2: median over plants of per-plant median (J_a0 - J) / J -0.01% [-0.09, +0.10], > 0 on 97 of 199; non-convergent 63 of 12800
5
+ x28m2b_s0_rep: median over plants of per-plant median (J_a0 - J) / J +1.92% [+0.98, +2.83], > 0 on 124 of 198; non-convergent 79 of 12736
6
+ x28m2b_s1_rep: median over plants of per-plant median (J_a0 - J) / J +2.08% [+1.09, +2.79], > 0 on 125 of 198; non-convergent 61 of 12736
7
+ x28m2b_s2_rep: median over plants of per-plant median (J_a0 - J) / J +2.06% [+0.62, +3.09], > 0 on 126 of 199; non-convergent 122 of 12800
8
+ x28m2c_s0: median over plants of per-plant median (J_a0 - J) / J +1.17% [+0.46, +2.24], > 0 on 137 of 199; non-convergent 0 of 12800
9
+ x28m2c_s1: median over plants of per-plant median (J_a0 - J) / J +1.12% [+0.43, +1.99], > 0 on 131 of 199; non-convergent 0 of 12800
10
+ x28m2c_s2: median over plants of per-plant median (J_a0 - J) / J +0.77% [+0.36, +1.35], > 0 on 132 of 199; non-convergent 0 of 12800
results/icml_r3/X28/mimo/fb_fresh_t2_m3.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ee4431d1b3c8caaadd234895fb9273f9cfae6d7ecbd28bdb210f4f5ef3e4a45
3
+ size 1754357
results/icml_r3/X28/mimo/fb_fresh_t2_m3.txt ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback, fresh directions on T2, m = 3: best certified b=16 sample per plant on 64 unit directions (seed (7979, plant index)), vs x28m3a_s0; _rep = (b) with repair, c = fallback arm
2
+ x28m3a_s0: median over plants of per-plant median (J_a0 - J) / J +0.00% [+0.00, +0.00], > 0 on 0 of 200; non-convergent 172 of 12800
3
+ x28m3a_s1: median over plants of per-plant median (J_a0 - J) / J +0.09% [-0.38, +0.44], > 0 on 102 of 200; non-convergent 200 of 12800
4
+ x28m3a_s2: median over plants of per-plant median (J_a0 - J) / J +0.05% [-0.21, +0.41], > 0 on 103 of 197; non-convergent 47 of 12608
5
+ x28m3b_s0_rep: median over plants of per-plant median (J_a0 - J) / J +1.72% [+0.86, +2.63], > 0 on 126 of 200; non-convergent 170 of 12800
6
+ x28m3b_s1_rep: median over plants of per-plant median (J_a0 - J) / J +2.39% [+1.42, +3.19], > 0 on 126 of 199; non-convergent 66 of 12800
7
+ x28m3b_s2_rep: median over plants of per-plant median (J_a0 - J) / J +1.55% [+0.39, +2.62], > 0 on 119 of 199; non-convergent 63 of 12736
8
+ x28m3c_s0: median over plants of per-plant median (J_a0 - J) / J +1.65% [+0.92, +2.62], > 0 on 137 of 200; non-convergent 117 of 12800
9
+ x28m3c_s1: median over plants of per-plant median (J_a0 - J) / J +2.19% [+1.53, +2.83], > 0 on 147 of 200; non-convergent 46 of 12800
10
+ x28m3c_s2: median over plants of per-plant median (J_a0 - J) / J +1.45% [+0.99, +2.13], > 0 on 134 of 199; non-convergent 98 of 12736
results/icml_r3/X28/mimo/fb_heldout_m2.txt ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback, m = 2, on the 120 HELD-OUT plants (they shaped the variant; secondary): (c) q = 8, c = 8 vs (a), regret vs J*
2
+ a s0: certified per sample 99.5% | b=1: certified 100.0% regret median 12.25% | b=4: certified 100.0% regret median 10.01% | b=16: certified 100.0% regret median 8.22%
3
+ a s1: certified per sample 98.8% | b=1: certified 98.3% regret median 10.18% | b=4: certified 99.2% regret median 8.49% | b=16: certified 100.0% regret median 7.95%
4
+ a s2: certified per sample 99.0% | b=1: certified 99.2% regret median 12.09% | b=4: certified 100.0% regret median 8.72% | b=16: certified 100.0% regret median 7.99%
5
+ b s0: certified per sample 94.1% | b=1: certified 86.7% regret median 11.65% | b=4: certified 95.0% regret median 6.25% | b=16: certified 100.0% regret median 4.51%
6
+ b s1: certified per sample 93.4% | b=1: certified 86.7% regret median 13.77% | b=4: certified 96.7% regret median 7.08% | b=16: certified 100.0% regret median 4.29%
7
+ b s2: certified per sample 92.0% | b=1: certified 79.2% regret median 14.90% | b=4: certified 94.2% regret median 7.02% | b=16: certified 100.0% regret median 5.00%
8
+ b= 1 seed 0: (a) - (b) regret median -2.01 pts [-3.81, +0.90] p_holm 0.019 | certified (b) - (a) -13.3 pts McNemar p 3.1e-05 | (b) cheaper on 44.2%, dearer on 55.8%
9
+ b= 1 seed 1: (a) - (b) regret median -3.54 pts [-10.03, +0.07] p_holm 0.00027 | certified (b) - (a) -11.7 pts McNemar p 0.0013 | (b) cheaper on 41.7%, dearer on 58.3%
10
+ b= 1 seed 2: (a) - (b) regret median -2.74 pts [-9.21, +0.94] p_holm 0.00027 | certified (b) - (a) -20.0 pts McNemar p 8e-07 | (b) cheaper on 45.8%, dearer on 54.2%
11
+ b= 4 seed 0: (a) - (b) regret median +1.48 pts [-0.26, +2.46] p_holm 0.93 | certified (b) - (a) -5.0 pts McNemar p 0.031 | (b) cheaper on 57.5%, dearer on 42.5%
12
+ b= 4 seed 1: (a) - (b) regret median +0.84 pts [-1.35, +2.28] p_holm 0.93 | certified (b) - (a) -2.5 pts McNemar p 0.38 | (b) cheaper on 55.8%, dearer on 44.2%
13
+ b= 4 seed 2: (a) - (b) regret median +0.32 pts [-1.43, +1.91] p_holm 0.62 | certified (b) - (a) -5.8 pts McNemar p 0.016 | (b) cheaper on 52.5%, dearer on 47.5%
14
+ b=16 seed 0: (a) - (b) regret median +2.05 pts [+1.25, +3.15] p_holm 2.5e-11 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 73.3%, dearer on 26.7%
15
+ b=16 seed 1: (a) - (b) regret median +1.34 pts [+0.43, +2.23] p_holm 1.5e-07 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 69.2%, dearer on 30.8%
16
+ b=16 seed 2: (a) - (b) regret median +1.38 pts [+0.89, +2.19] p_holm 5.4e-11 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 75.0%, dearer on 25.0%
17
+ VERDICT m=2 b=16: better on 3, worse on 0 of 3 seed pairs -> spline better
18
+ tails, m = 2: d = (a) - (c) at b = 16 (points)
19
+ s0: median +2.05, mean +3.76; (c) better by > 5 on 37, worse by > 5 on 1, by > 10 on 1; Wilcoxon p before Holm 8.49e-12; fallback on 100.0%
20
+ (c) worse by > 10 n = 1: spline samples certified as sampled (of 16) median 0.0; fallback on 100.0%; median regret (c) 59.2%, (a) 39.0%
21
+ (c) better by > 5 n = 37: spline samples certified as sampled (of 16) median 16.0; fallback on 100.0%; median regret (c) 4.2%, (a) 13.2%
22
+ all plants n = 120: spline samples certified as sampled (of 16) median 15.0; fallback on 100.0%; median regret (c) 4.5%, (a) 8.2%
23
+ s1: median +1.34, mean +3.78; (c) better by > 5 on 34, worse by > 5 on 6, by > 10 on 2; Wilcoxon p before Holm 1.47e-07; fallback on 100.0%
24
+ (c) worse by > 10 n = 2: spline samples certified as sampled (of 16) median 6.0; fallback on 100.0%; median regret (c) 63.6%, (a) 19.6%
25
+ (c) better by > 5 n = 34: spline samples certified as sampled (of 16) median 15.5; fallback on 100.0%; median regret (c) 3.2%, (a) 13.0%
26
+ all plants n = 120: spline samples certified as sampled (of 16) median 15.0; fallback on 100.0%; median regret (c) 4.3%, (a) 7.9%
27
+ s2: median +1.38, mean +3.22; (c) better by > 5 on 33, worse by > 5 on 2, by > 10 on 1; Wilcoxon p before Holm 2.69e-11; fallback on 100.0%
28
+ (c) worse by > 10 n = 1: spline samples certified as sampled (of 16) median 7.0; fallback on 100.0%; median regret (c) 30.1%, (a) 9.6%
29
+ (c) better by > 5 n = 33: spline samples certified as sampled (of 16) median 16.0; fallback on 100.0%; median regret (c) 4.4%, (a) 13.3%
30
+ all plants n = 120: spline samples certified as sampled (of 16) median 15.0; fallback on 100.0%; median regret (c) 5.0%, (a) 8.0%
results/icml_r3/X28/mimo/fb_heldout_m3.txt ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback, m = 3, on the 120 HELD-OUT plants (they shaped the variant; secondary): (c) q = 8, c = 8 vs (a), regret vs J*
2
+ a s0: certified per sample 96.0% | b=1: certified 95.0% regret median 17.17% | b=4: certified 100.0% regret median 13.51% | b=16: certified 100.0% regret median 10.66%
3
+ a s1: certified per sample 96.6% | b=1: certified 97.5% regret median 16.83% | b=4: certified 100.0% regret median 11.77% | b=16: certified 100.0% regret median 10.46%
4
+ a s2: certified per sample 93.4% | b=1: certified 94.2% regret median 16.33% | b=4: certified 97.5% regret median 12.10% | b=16: certified 100.0% regret median 10.56%
5
+ b s0: certified per sample 87.8% | b=1: certified 83.3% regret median 23.66% | b=4: certified 96.7% regret median 11.02% | b=16: certified 100.0% regret median 7.63%
6
+ b s1: certified per sample 88.0% | b=1: certified 80.0% regret median 23.94% | b=4: certified 95.0% regret median 11.72% | b=16: certified 100.0% regret median 8.00%
7
+ b s2: certified per sample 88.1% | b=1: certified 80.8% regret median 23.89% | b=4: certified 99.2% regret median 13.16% | b=16: certified 100.0% regret median 8.42%
8
+ b= 1 seed 0: (a) - (b) regret median -3.10 pts [-8.25, +0.31] p_holm 0.006 | certified (b) - (a) -11.7 pts McNemar p 0.0043 | (b) cheaper on 41.7%, dearer on 56.7%
9
+ b= 1 seed 1: (a) - (b) regret median -4.18 pts [-9.36, +0.04] p_holm 0.00071 | certified (b) - (a) -17.5 pts McNemar p 1.9e-05 | (b) cheaper on 40.8%, dearer on 58.3%
10
+ b= 1 seed 2: (a) - (b) regret median -2.14 pts [-6.33, +0.15] p_holm 0.006 | certified (b) - (a) -13.3 pts McNemar p 0.0015 | (b) cheaper on 42.5%, dearer on 55.0%
11
+ b= 4 seed 0: (a) - (b) regret median +0.62 pts [-1.47, +2.24] p_holm 1 | certified (b) - (a) -3.3 pts McNemar p 0.12 | (b) cheaper on 52.5%, dearer on 47.5%
12
+ b= 4 seed 1: (a) - (b) regret median +0.45 pts [-0.97, +2.41] p_holm 1 | certified (b) - (a) -5.0 pts McNemar p 0.031 | (b) cheaper on 55.0%, dearer on 45.0%
13
+ b= 4 seed 2: (a) - (b) regret median +1.07 pts [-0.80, +2.78] p_holm 1 | certified (b) - (a) +1.7 pts McNemar p 0.62 | (b) cheaper on 53.3%, dearer on 46.7%
14
+ b=16 seed 0: (a) - (b) regret median +1.18 pts [+0.56, +2.02] p_holm 1.5e-05 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 66.7%, dearer on 33.3%
15
+ b=16 seed 1: (a) - (b) regret median +1.46 pts [+0.30, +2.39] p_holm 1.5e-05 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 65.0%, dearer on 35.0%
16
+ b=16 seed 2: (a) - (b) regret median +1.33 pts [+0.33, +2.98] p_holm 1.7e-06 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 65.0%, dearer on 35.0%
17
+ VERDICT m=3 b=16: better on 3, worse on 0 of 3 seed pairs -> spline better
18
+ tails, m = 3: d = (a) - (c) at b = 16 (points)
19
+ s0: median +1.18, mean +2.95; (c) better by > 5 on 27, worse by > 5 on 4, by > 10 on 1; Wilcoxon p before Holm 1.07e-05; fallback on 100.0%
20
+ (c) worse by > 10 n = 1: spline samples certified as sampled (of 16) median 1.0; fallback on 100.0%; median regret (c) 34.5%, (a) 23.7%
21
+ (c) better by > 5 n = 27: spline samples certified as sampled (of 16) median 12.0; fallback on 100.0%; median regret (c) 4.3%, (a) 13.5%
22
+ all plants n = 120: spline samples certified as sampled (of 16) median 10.0; fallback on 100.0%; median regret (c) 7.6%, (a) 10.7%
23
+ s1: median +1.46, mean +3.47; (c) better by > 5 on 30, worse by > 5 on 4, by > 10 on 1; Wilcoxon p before Holm 7.34e-06; fallback on 100.0%
24
+ (c) worse by > 10 n = 1: spline samples certified as sampled (of 16) median 8.0; fallback on 100.0%; median regret (c) 25.4%, (a) 13.0%
25
+ (c) better by > 5 n = 30: spline samples certified as sampled (of 16) median 12.5; fallback on 100.0%; median regret (c) 5.5%, (a) 15.5%
26
+ all plants n = 120: spline samples certified as sampled (of 16) median 10.0; fallback on 100.0%; median regret (c) 8.0%, (a) 10.5%
27
+ s2: median +1.33, mean +6.06; (c) better by > 5 on 39, worse by > 5 on 4, by > 10 on 0; Wilcoxon p before Holm 5.75e-07; fallback on 100.0%
28
+ (c) better by > 5 n = 39: spline samples certified as sampled (of 16) median 11.0; fallback on 100.0%; median regret (c) 6.5%, (a) 17.6%
29
+ all plants n = 120: spline samples certified as sampled (of 16) median 10.0; fallback on 100.0%; median regret (c) 8.4%, (a) 10.6%
results/icml_r3/X28/mimo/fb_paired_t2_m2.txt ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback, m = 2: PRIMARY, (c) spline with identity fallback (q = 8, c = 8) vs (a), 200 fresh T2 plants, regret vs J*_T2 (b = 1, 4 of (c) equal (b) with repair)
2
+ a s0: certified per sample 98.6% | b=1: certified 98.5% regret median 9.19% | b=4: certified 99.0% regret median 7.28% | b=16: certified 99.5% regret median 6.39%
3
+ a s1: certified per sample 98.7% | b=1: certified 98.5% regret median 8.87% | b=4: certified 99.5% regret median 7.49% | b=16: certified 100.0% regret median 6.47%
4
+ a s2: certified per sample 98.4% | b=1: certified 98.5% regret median 8.82% | b=4: certified 99.5% regret median 7.14% | b=16: certified 100.0% regret median 6.39%
5
+ b s0: certified per sample 94.3% | b=1: certified 91.0% regret median 10.29% | b=4: certified 98.5% regret median 5.04% | b=16: certified 100.0% regret median 3.33%
6
+ b s1: certified per sample 94.1% | b=1: certified 90.0% regret median 9.85% | b=4: certified 97.5% regret median 5.49% | b=16: certified 100.0% regret median 3.54%
7
+ b s2: certified per sample 94.3% | b=1: certified 90.5% regret median 10.38% | b=4: certified 97.5% regret median 5.89% | b=16: certified 100.0% regret median 3.40%
8
+ b= 1 seed 0: (a) - (b) regret median +0.06 pts [-1.82, +1.21] p_holm 0.02 | certified (b) - (a) -7.5 pts McNemar p 0.00073 | (b) cheaper on 50.5%, dearer on 49.0%
9
+ b= 1 seed 1: (a) - (b) regret median -1.33 pts [-2.71, +0.00] p_holm 0.00045 | certified (b) - (a) -8.5 pts McNemar p 7.6e-05 | (b) cheaper on 42.5%, dearer on 56.5%
10
+ b= 1 seed 2: (a) - (b) regret median -1.11 pts [-3.24, +0.23] p_holm 0.00056 | certified (b) - (a) -8.0 pts McNemar p 0.00014 | (b) cheaper on 44.5%, dearer on 54.5%
11
+ b= 4 seed 0: (a) - (b) regret median +1.07 pts [+0.01, +2.43] p_holm 0.76 | certified (b) - (a) -0.5 pts McNemar p 1 | (b) cheaper on 57.5%, dearer on 42.5%
12
+ b= 4 seed 1: (a) - (b) regret median +0.89 pts [-0.20, +1.99] p_holm 1 | certified (b) - (a) -2.0 pts McNemar p 0.22 | (b) cheaper on 55.5%, dearer on 44.5%
13
+ b= 4 seed 2: (a) - (b) regret median +0.18 pts [-1.21, +1.96] p_holm 1 | certified (b) - (a) -2.0 pts McNemar p 0.12 | (b) cheaper on 50.5%, dearer on 49.0%
14
+ b=16 seed 0: (a) - (b) regret median +1.69 pts [+0.90, +2.93] p_holm 4e-16 | certified (b) - (a) +0.5 pts McNemar p 1 | (b) cheaper on 73.5%, dearer on 26.5%
15
+ b=16 seed 1: (a) - (b) regret median +0.98 pts [+0.59, +2.14] p_holm 5.5e-12 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 70.0%, dearer on 30.0%
16
+ b=16 seed 2: (a) - (b) regret median +1.04 pts [+0.40, +2.25] p_holm 1.4e-11 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 65.5%, dearer on 34.5%
17
+ VERDICT m=2 b=16: better on 2, worse on 0 of 3 seed pairs -> spline better
18
+ fallback on s0 100.0%, s1 100.0%, s2 100.0% of the plants
19
+
20
+ tails, m = 2: d = (a) - (c) at b = 16 (points)
21
+ s0: median +1.69, mean +2.95; (c) better by > 5 on 57, worse by > 5 on 1, by > 10 on 1; Wilcoxon p before Holm 1.35e-16; fallback on 100.0%
22
+ (c) worse by > 10 n = 1: spline samples certified as sampled (of 16) median 0.0; fallback on 100.0%; median regret (c) 87.1%, (a) 70.6%
23
+ (c) better by > 5 n = 57: spline samples certified as sampled (of 16) median 14.0; fallback on 100.0%; median regret (c) 2.3%, (a) 11.1%
24
+ all plants n = 200: spline samples certified as sampled (of 16) median 14.0; fallback on 100.0%; median regret (c) 3.3%, (a) 6.4%
25
+ s1: median +0.98, mean +1.91; (c) better by > 5 on 49, worse by > 5 on 6, by > 10 on 3; Wilcoxon p before Holm 2.74e-12; fallback on 100.0%
26
+ (c) worse by > 10 n = 3: spline samples certified as sampled (of 16) median 0.0; fallback on 100.0%; median regret (c) 89.3%, (a) 23.3%
27
+ (c) better by > 5 n = 49: spline samples certified as sampled (of 16) median 16.0; fallback on 100.0%; median regret (c) 2.4%, (a) 11.1%
28
+ all plants n = 200: spline samples certified as sampled (of 16) median 15.0; fallback on 100.0%; median regret (c) 3.5%, (a) 6.5%
29
+ s2: median +1.04, mean +1.86; (c) better by > 5 on 53, worse by > 5 on 5, by > 10 on 3; Wilcoxon p before Holm 1.42e-11; fallback on 100.0%
30
+ (c) worse by > 10 n = 3: spline samples certified as sampled (of 16) median 4.0; fallback on 100.0%; median regret (c) 44.1%, (a) 27.8%
31
+ (c) better by > 5 n = 53: spline samples certified as sampled (of 16) median 15.0; fallback on 100.0%; median regret (c) 2.3%, (a) 11.1%
32
+ all plants n = 200: spline samples certified as sampled (of 16) median 15.0; fallback on 100.0%; median regret (c) 3.4%, (a) 6.4%
33
+
34
+ replication (secondary): (b) with repair R vs (a) (the 2026-11-14 test) on T2
35
+ a s0: certified per sample 98.6% | b=1: certified 98.5% regret median 9.19% | b=4: certified 99.0% regret median 7.28% | b=16: certified 99.5% regret median 6.39%
36
+ a s1: certified per sample 98.7% | b=1: certified 98.5% regret median 8.87% | b=4: certified 99.5% regret median 7.49% | b=16: certified 100.0% regret median 6.47%
37
+ a s2: certified per sample 98.4% | b=1: certified 98.5% regret median 8.82% | b=4: certified 99.5% regret median 7.14% | b=16: certified 100.0% regret median 6.39%
38
+ b s0: certified per sample 89.5% | b=1: certified 91.0% regret median 10.29% | b=4: certified 98.5% regret median 5.04% | b=16: certified 99.5% regret median 3.51%
39
+ b s1: certified per sample 89.6% | b=1: certified 90.0% regret median 9.85% | b=4: certified 97.5% regret median 5.49% | b=16: certified 99.5% regret median 3.62%
40
+ b s2: certified per sample 90.8% | b=1: certified 90.5% regret median 10.38% | b=4: certified 97.5% regret median 5.89% | b=16: certified 100.0% regret median 3.64%
41
+ b= 1 seed 0: (a) - (b) regret median +0.06 pts [-1.82, +1.21] p_holm 0.02 | certified (b) - (a) -7.5 pts McNemar p 0.00073 | (b) cheaper on 50.5%, dearer on 49.0%
42
+ b= 1 seed 1: (a) - (b) regret median -1.33 pts [-2.71, +0.00] p_holm 0.00045 | certified (b) - (a) -8.5 pts McNemar p 7.6e-05 | (b) cheaper on 42.5%, dearer on 56.5%
43
+ b= 1 seed 2: (a) - (b) regret median -1.11 pts [-3.24, +0.23] p_holm 0.00056 | certified (b) - (a) -8.0 pts McNemar p 0.00014 | (b) cheaper on 44.5%, dearer on 54.5%
44
+ b= 4 seed 0: (a) - (b) regret median +1.07 pts [+0.01, +2.43] p_holm 0.76 | certified (b) - (a) -0.5 pts McNemar p 1 | (b) cheaper on 57.5%, dearer on 42.5%
45
+ b= 4 seed 1: (a) - (b) regret median +0.89 pts [-0.20, +1.99] p_holm 1 | certified (b) - (a) -2.0 pts McNemar p 0.22 | (b) cheaper on 55.5%, dearer on 44.5%
46
+ b= 4 seed 2: (a) - (b) regret median +0.18 pts [-1.21, +1.96] p_holm 1 | certified (b) - (a) -2.0 pts McNemar p 0.12 | (b) cheaper on 50.5%, dearer on 49.0%
47
+ b=16 seed 0: (a) - (b) regret median +2.33 pts [+1.14, +3.27] p_holm 0.00029 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 64.5%, dearer on 35.5%
48
+ b=16 seed 1: (a) - (b) regret median +1.76 pts [+0.88, +3.25] p_holm 0.0019 | certified (b) - (a) -0.5 pts McNemar p 1 | (b) cheaper on 63.0%, dearer on 37.0%
49
+ b=16 seed 2: (a) - (b) regret median +1.78 pts [+0.78, +3.16] p_holm 0.0017 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 63.5%, dearer on 36.5%
50
+ VERDICT m=2 b=16: better on 3, worse on 0 of 3 seed pairs -> spline better
51
+
52
+ replication (secondary): (b) as sampled vs (a) (the 2026-11-13 test) on T2
53
+ a s0: certified per sample 98.6% | b=1: certified 98.5% regret median 9.19% | b=4: certified 99.0% regret median 7.28% | b=16: certified 99.5% regret median 6.39%
54
+ a s1: certified per sample 98.7% | b=1: certified 98.5% regret median 8.87% | b=4: certified 99.5% regret median 7.49% | b=16: certified 100.0% regret median 6.47%
55
+ a s2: certified per sample 98.4% | b=1: certified 98.5% regret median 8.82% | b=4: certified 99.5% regret median 7.14% | b=16: certified 100.0% regret median 6.39%
56
+ b s0: certified per sample 76.9% | b=1: certified 78.0% regret median 10.80% | b=4: certified 91.0% regret median 5.32% | b=16: certified 96.5% regret median 3.59%
57
+ b s1: certified per sample 76.5% | b=1: certified 75.5% regret median 10.17% | b=4: certified 92.0% regret median 5.49% | b=16: certified 96.5% regret median 3.66%
58
+ b s2: certified per sample 78.7% | b=1: certified 80.5% regret median 11.08% | b=4: certified 95.0% regret median 6.13% | b=16: certified 98.5% regret median 3.70%
59
+ b= 1 seed 0: (a) - (b) regret median -0.86 pts [-3.53, +0.22] p_holm 3.8e-05 | certified (b) - (a) -20.5 pts McNemar p 9.1e-13 | (b) cheaper on 45.0%, dearer on 53.5%
60
+ b= 1 seed 1: (a) - (b) regret median -2.42 pts [-5.51, -0.92] p_holm 3.4e-07 | certified (b) - (a) -23.0 pts McNemar p 3.5e-13 | (b) cheaper on 39.0%, dearer on 60.0%
61
+ b= 1 seed 2: (a) - (b) regret median -2.24 pts [-5.50, -0.15] p_holm 2e-06 | certified (b) - (a) -18.0 pts McNemar p 2.8e-10 | (b) cheaper on 41.0%, dearer on 58.0%
62
+ b= 4 seed 0: (a) - (b) regret median +0.48 pts [-0.58, +1.62] p_holm 1 | certified (b) - (a) -8.0 pts McNemar p 3.1e-05 | (b) cheaper on 53.5%, dearer on 45.5%
63
+ b= 4 seed 1: (a) - (b) regret median +0.51 pts [-0.53, +1.45] p_holm 1 | certified (b) - (a) -7.5 pts McNemar p 6.1e-05 | (b) cheaper on 53.0%, dearer on 46.5%
64
+ b= 4 seed 2: (a) - (b) regret median -0.10 pts [-1.72, +1.81] p_holm 1 | certified (b) - (a) -4.5 pts McNemar p 0.0039 | (b) cheaper on 48.5%, dearer on 51.0%
65
+ b=16 seed 0: (a) - (b) regret median +1.74 pts [+0.85, +3.04] p_holm 0.0093 | certified (b) - (a) -3.0 pts McNemar p 0.07 | (b) cheaper on 62.5%, dearer on 37.5%
66
+ b=16 seed 1: (a) - (b) regret median +1.57 pts [+0.69, +3.08] p_holm 0.018 | certified (b) - (a) -3.5 pts McNemar p 0.016 | (b) cheaper on 61.5%, dearer on 38.5%
67
+ b=16 seed 2: (a) - (b) regret median +1.63 pts [+0.64, +3.16] p_holm 0.018 | certified (b) - (a) -1.5 pts McNemar p 0.25 | (b) cheaper on 61.5%, dearer on 38.5%
68
+ VERDICT m=2 b=16: better on 1, worse on 0 of 3 seed pairs -> no measurable difference
results/icml_r3/X28/mimo/fb_paired_t2_m3.txt ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback, m = 3: PRIMARY, (c) spline with identity fallback (q = 8, c = 8) vs (a), 200 fresh T2 plants, regret vs J*_T2 (b = 1, 4 of (c) equal (b) with repair)
2
+ a s0: certified per sample 95.4% | b=1: certified 95.5% regret median 16.04% | b=4: certified 99.5% regret median 12.72% | b=16: certified 100.0% regret median 10.05%
3
+ a s1: certified per sample 95.8% | b=1: certified 97.5% regret median 16.23% | b=4: certified 100.0% regret median 13.39% | b=16: certified 100.0% regret median 10.24%
4
+ a s2: certified per sample 93.7% | b=1: certified 94.0% regret median 16.69% | b=4: certified 98.0% regret median 12.08% | b=16: certified 98.5% regret median 10.36%
5
+ b s0: certified per sample 88.0% | b=1: certified 80.5% regret median 18.21% | b=4: certified 96.0% regret median 9.47% | b=16: certified 100.0% regret median 6.20%
6
+ b s1: certified per sample 87.7% | b=1: certified 78.0% regret median 16.17% | b=4: certified 93.0% regret median 9.73% | b=16: certified 100.0% regret median 6.00%
7
+ b s2: certified per sample 86.2% | b=1: certified 80.5% regret median 18.82% | b=4: certified 92.5% regret median 9.91% | b=16: certified 99.5% regret median 6.25%
8
+ b= 1 seed 0: (a) - (b) regret median -1.35 pts [-4.20, +0.82] p_holm 0.0025 | certified (b) - (a) -15.0 pts McNemar p 6e-07 | (b) cheaper on 45.0%, dearer on 52.5%
9
+ b= 1 seed 1: (a) - (b) regret median -0.65 pts [-4.15, +1.84] p_holm 0.0025 | certified (b) - (a) -19.5 pts McNemar p 8.7e-10 | (b) cheaper on 47.0%, dearer on 52.0%
10
+ b= 1 seed 2: (a) - (b) regret median -0.86 pts [-3.30, +0.79] p_holm 0.0045 | certified (b) - (a) -13.5 pts McNemar p 2.5e-05 | (b) cheaper on 46.0%, dearer on 51.5%
11
+ b= 4 seed 0: (a) - (b) regret median +2.35 pts [+0.26, +4.58] p_holm 1 | certified (b) - (a) -3.5 pts McNemar p 0.016 | (b) cheaper on 58.0%, dearer on 41.5%
12
+ b= 4 seed 1: (a) - (b) regret median +2.26 pts [-0.13, +3.92] p_holm 1 | certified (b) - (a) -7.0 pts McNemar p 0.00012 | (b) cheaper on 55.5%, dearer on 44.5%
13
+ b= 4 seed 2: (a) - (b) regret median +1.77 pts [-0.06, +3.90] p_holm 1 | certified (b) - (a) -5.5 pts McNemar p 0.0074 | (b) cheaper on 56.0%, dearer on 43.0%
14
+ b=16 seed 0: (a) - (b) regret median +3.03 pts [+1.80, +4.37] p_holm 1.8e-14 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 73.0%, dearer on 27.0%
15
+ b=16 seed 1: (a) - (b) regret median +2.40 pts [+1.47, +3.83] p_holm 7.5e-14 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 72.0%, dearer on 28.0%
16
+ b=16 seed 2: (a) - (b) regret median +2.43 pts [+1.13, +3.90] p_holm 2.7e-11 | certified (b) - (a) +1.0 pts McNemar p 0.5 | (b) cheaper on 70.0%, dearer on 29.5%
17
+ VERDICT m=3 b=16: better on 3, worse on 0 of 3 seed pairs -> spline better
18
+ fallback on s0 100.0%, s1 100.0%, s2 100.0% of the plants
19
+
20
+ tails, m = 3: d = (a) - (c) at b = 16 (points)
21
+ s0: median +3.03, mean +4.68; (c) better by > 5 on 77, worse by > 5 on 11, by > 10 on 5; Wilcoxon p before Holm 5.93e-15; fallback on 100.0%
22
+ (c) worse by > 10 n = 5: spline samples certified as sampled (of 16) median 1.0; fallback on 100.0%; median regret (c) 60.9%, (a) 41.4%
23
+ (c) better by > 5 n = 77: spline samples certified as sampled (of 16) median 12.0; fallback on 100.0%; median regret (c) 4.6%, (a) 14.0%
24
+ all plants n = 200: spline samples certified as sampled (of 16) median 11.0; fallback on 100.0%; median regret (c) 6.2%, (a) 10.0%
25
+ s1: median +2.40, mean +4.27; (c) better by > 5 on 69, worse by > 5 on 9, by > 10 on 5; Wilcoxon p before Holm 3.73e-14; fallback on 100.0%
26
+ (c) worse by > 10 n = 5: spline samples certified as sampled (of 16) median 4.0; fallback on 100.0%; median regret (c) 28.5%, (a) 17.4%
27
+ (c) better by > 5 n = 69: spline samples certified as sampled (of 16) median 13.0; fallback on 100.0%; median regret (c) 4.2%, (a) 14.3%
28
+ all plants n = 200: spline samples certified as sampled (of 16) median 11.0; fallback on 100.0%; median regret (c) 6.0%, (a) 10.2%
29
+ s2: median +2.43, mean +3.08; (c) better by > 5 on 74, worse by > 5 on 15, by > 10 on 6; Wilcoxon p before Holm 2.75e-11; fallback on 100.0%
30
+ (c) worse by > 10 n = 6: spline samples certified as sampled (of 16) median 1.5; fallback on 100.0%; median regret (c) 61.8%, (a) 31.0%
31
+ (c) better by > 5 n = 74: spline samples certified as sampled (of 16) median 12.0; fallback on 100.0%; median regret (c) 5.1%, (a) 14.6%
32
+ all plants n = 200: spline samples certified as sampled (of 16) median 11.0; fallback on 100.0%; median regret (c) 6.2%, (a) 10.4%
33
+
34
+ replication (secondary): (b) with repair R vs (a) (the 2026-11-14 test) on T2
35
+ a s0: certified per sample 95.4% | b=1: certified 95.5% regret median 16.04% | b=4: certified 99.5% regret median 12.72% | b=16: certified 100.0% regret median 10.05%
36
+ a s1: certified per sample 95.8% | b=1: certified 97.5% regret median 16.23% | b=4: certified 100.0% regret median 13.39% | b=16: certified 100.0% regret median 10.24%
37
+ a s2: certified per sample 93.7% | b=1: certified 94.0% regret median 16.69% | b=4: certified 98.0% regret median 12.08% | b=16: certified 98.5% regret median 10.36%
38
+ b s0: certified per sample 81.5% | b=1: certified 80.5% regret median 18.21% | b=4: certified 96.0% regret median 9.47% | b=16: certified 100.0% regret median 6.16%
39
+ b s1: certified per sample 79.3% | b=1: certified 78.0% regret median 16.17% | b=4: certified 93.0% regret median 9.73% | b=16: certified 100.0% regret median 6.82%
40
+ b s2: certified per sample 78.8% | b=1: certified 80.5% regret median 18.82% | b=4: certified 92.5% regret median 9.91% | b=16: certified 99.5% regret median 6.71%
41
+ b= 1 seed 0: (a) - (b) regret median -1.35 pts [-4.20, +0.82] p_holm 0.0025 | certified (b) - (a) -15.0 pts McNemar p 6e-07 | (b) cheaper on 45.0%, dearer on 52.5%
42
+ b= 1 seed 1: (a) - (b) regret median -0.65 pts [-4.15, +1.84] p_holm 0.0025 | certified (b) - (a) -19.5 pts McNemar p 8.7e-10 | (b) cheaper on 47.0%, dearer on 52.0%
43
+ b= 1 seed 2: (a) - (b) regret median -0.86 pts [-3.30, +0.79] p_holm 0.0045 | certified (b) - (a) -13.5 pts McNemar p 2.5e-05 | (b) cheaper on 46.0%, dearer on 51.5%
44
+ b= 4 seed 0: (a) - (b) regret median +2.35 pts [+0.26, +4.58] p_holm 1 | certified (b) - (a) -3.5 pts McNemar p 0.016 | (b) cheaper on 58.0%, dearer on 41.5%
45
+ b= 4 seed 1: (a) - (b) regret median +2.26 pts [-0.13, +3.92] p_holm 1 | certified (b) - (a) -7.0 pts McNemar p 0.00012 | (b) cheaper on 55.5%, dearer on 44.5%
46
+ b= 4 seed 2: (a) - (b) regret median +1.77 pts [-0.06, +3.90] p_holm 1 | certified (b) - (a) -5.5 pts McNemar p 0.0074 | (b) cheaper on 56.0%, dearer on 43.0%
47
+ b=16 seed 0: (a) - (b) regret median +3.40 pts [+2.20, +4.38] p_holm 0.00068 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 66.5%, dearer on 33.5%
48
+ b=16 seed 1: (a) - (b) regret median +3.11 pts [+1.94, +4.71] p_holm 0.00074 | certified (b) - (a) +0.0 pts McNemar p 1 | (b) cheaper on 65.5%, dearer on 34.5%
49
+ b=16 seed 2: (a) - (b) regret median +3.72 pts [+1.55, +5.17] p_holm 0.00074 | certified (b) - (a) +1.0 pts McNemar p 0.62 | (b) cheaper on 64.0%, dearer on 36.0%
50
+ VERDICT m=3 b=16: better on 3, worse on 0 of 3 seed pairs -> spline better
51
+
52
+ replication (secondary): (b) as sampled vs (a) (the 2026-11-13 test) on T2
53
+ a s0: certified per sample 95.4% | b=1: certified 95.5% regret median 16.04% | b=4: certified 99.5% regret median 12.72% | b=16: certified 100.0% regret median 10.05%
54
+ a s1: certified per sample 95.8% | b=1: certified 97.5% regret median 16.23% | b=4: certified 100.0% regret median 13.39% | b=16: certified 100.0% regret median 10.24%
55
+ a s2: certified per sample 93.7% | b=1: certified 94.0% regret median 16.69% | b=4: certified 98.0% regret median 12.08% | b=16: certified 98.5% regret median 10.36%
56
+ b s0: certified per sample 62.0% | b=1: certified 62.0% regret median 27.16% | b=4: certified 85.0% regret median 9.91% | b=16: certified 95.5% regret median 6.16%
57
+ b s1: certified per sample 60.5% | b=1: certified 61.0% regret median 16.89% | b=4: certified 81.5% regret median 9.78% | b=16: certified 95.0% regret median 6.90%
58
+ b s2: certified per sample 61.5% | b=1: certified 61.5% regret median 24.51% | b=4: certified 81.0% regret median 10.21% | b=16: certified 95.5% regret median 6.74%
59
+ b= 1 seed 0: (a) - (b) regret median -2.34 pts [-12.62, +0.00] p_holm 2.4e-07 | certified (b) - (a) -33.5 pts McNemar p 2.2e-18 | (b) cheaper on 40.5%, dearer on 56.0%
60
+ b= 1 seed 1: (a) - (b) regret median -2.83 pts [-9.74, +0.00] p_holm 3.1e-07 | certified (b) - (a) -36.5 pts McNemar p 4e-21 | (b) cheaper on 42.0%, dearer on 56.0%
61
+ b= 1 seed 2: (a) - (b) regret median -2.97 pts [-11.69, -0.70] p_holm 1.9e-07 | certified (b) - (a) -32.5 pts McNemar p 2.4e-16 | (b) cheaper on 38.5%, dearer on 57.5%
62
+ b= 4 seed 0: (a) - (b) regret median +0.75 pts [-0.81, +2.86] p_holm 0.34 | certified (b) - (a) -14.5 pts McNemar p 3.7e-09 | (b) cheaper on 53.5%, dearer on 46.0%
63
+ b= 4 seed 1: (a) - (b) regret median +0.79 pts [-0.54, +2.86] p_holm 0.34 | certified (b) - (a) -18.5 pts McNemar p 1.5e-11 | (b) cheaper on 53.0%, dearer on 47.0%
64
+ b= 4 seed 2: (a) - (b) regret median +1.48 pts [-0.40, +3.38] p_holm 0.35 | certified (b) - (a) -17.0 pts McNemar p 1.1e-09 | (b) cheaper on 54.0%, dearer on 44.5%
65
+ b=16 seed 0: (a) - (b) regret median +3.00 pts [+1.42, +4.22] p_holm 0.062 | certified (b) - (a) -4.5 pts McNemar p 0.0039 | (b) cheaper on 63.0%, dearer on 37.0%
66
+ b=16 seed 1: (a) - (b) regret median +2.63 pts [+1.48, +4.15] p_holm 0.063 | certified (b) - (a) -5.0 pts McNemar p 0.002 | (b) cheaper on 62.5%, dearer on 37.5%
67
+ b=16 seed 2: (a) - (b) regret median +2.86 pts [+0.94, +5.10] p_holm 0.062 | certified (b) - (a) -3.0 pts McNemar p 0.11 | (b) cheaper on 61.5%, dearer on 38.0%
68
+ VERDICT m=3 b=16: better on 0, worse on 0 of 3 seed pairs -> no measurable difference
results/icml_r3/X28/mimo/fb_val_choice.txt ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ X28-M fallback: choice of (q, c) on the 64 validation plants per m (PREREG addendum 2026-11-15), b = 16, pooled over the 3 seed pairs
2
+ d = regret (a) - regret (c); z = standardized Wilcoxon signed-rank statistic of d; eligible iff certified (c) >= certified (a) - 2 points
3
+ m = 3: arm (a) certified 100.0%
4
+ q = 4 c = -1: z +4.44, median d +2.82 pts, certified 99.0%, fallback on 0.0% of plants
5
+ q = 4 c = 0: z +5.52, median d +2.76 pts, certified 100.0%, fallback on 15.6% of plants
6
+ q = 4 c = 1: z +6.51, median d +2.55 pts, certified 100.0%, fallback on 35.9% of plants
7
+ q = 4 c = 2: z +6.88, median d +2.52 pts, certified 100.0%, fallback on 46.4% of plants
8
+ q = 4 c = 3: z +7.13, median d +2.22 pts, certified 100.0%, fallback on 65.6% of plants
9
+ q = 4 c = 4: z +7.36, median d +1.83 pts, certified 100.0%, fallback on 100.0% of plants
10
+ q = 8 c = -1: z +4.44, median d +2.82 pts, certified 99.0%, fallback on 0.0% of plants
11
+ q = 8 c = 0: z +5.39, median d +2.82 pts, certified 100.0%, fallback on 10.4% of plants
12
+ q = 8 c = 1: z +5.88, median d +2.82 pts, certified 100.0%, fallback on 16.7% of plants
13
+ q = 8 c = 2: z +6.67, median d +2.87 pts, certified 100.0%, fallback on 25.0% of plants
14
+ q = 8 c = 3: z +7.24, median d +2.87 pts, certified 100.0%, fallback on 34.4% of plants
15
+ q = 8 c = 4: z +7.24, median d +2.63 pts, certified 100.0%, fallback on 42.7% of plants
16
+ q = 8 c = 5: z +7.12, median d +2.55 pts, certified 100.0%, fallback on 51.0% of plants
17
+ q = 8 c = 6: z +7.27, median d +2.52 pts, certified 100.0%, fallback on 58.9% of plants
18
+ q = 8 c = 7: z +7.42, median d +2.49 pts, certified 100.0%, fallback on 76.0% of plants
19
+ q = 8 c = 8: z +7.63, median d +2.45 pts, certified 100.0%, fallback on 100.0% of plants
20
+ CHOICE m = 3: q = 8, c = 8
21
+ m = 2: arm (a) certified 100.0%
22
+ q = 4 c = -1: z +3.91, median d +2.46 pts, certified 100.0%, fallback on 0.0% of plants
23
+ q = 4 c = 0: z +4.60, median d +2.46 pts, certified 100.0%, fallback on 7.3% of plants
24
+ q = 4 c = 1: z +5.13, median d +2.43 pts, certified 100.0%, fallback on 14.1% of plants
25
+ q = 4 c = 2: z +5.62, median d +1.92 pts, certified 100.0%, fallback on 27.6% of plants
26
+ q = 4 c = 3: z +6.70, median d +1.74 pts, certified 100.0%, fallback on 44.8% of plants
27
+ q = 4 c = 4: z +7.66, median d +1.11 pts, certified 100.0%, fallback on 100.0% of plants
28
+ q = 8 c = -1: z +3.91, median d +2.46 pts, certified 100.0%, fallback on 0.0% of plants
29
+ q = 8 c = 0: z +4.59, median d +2.52 pts, certified 100.0%, fallback on 4.7% of plants
30
+ q = 8 c = 1: z +4.63, median d +2.52 pts, certified 100.0%, fallback on 7.8% of plants
31
+ q = 8 c = 2: z +4.61, median d +2.52 pts, certified 100.0%, fallback on 8.9% of plants
32
+ q = 8 c = 3: z +5.34, median d +2.52 pts, certified 100.0%, fallback on 15.6% of plants
33
+ q = 8 c = 4: z +5.47, median d +2.52 pts, certified 100.0%, fallback on 18.8% of plants
34
+ q = 8 c = 5: z +6.07, median d +2.26 pts, certified 100.0%, fallback on 31.2% of plants
35
+ q = 8 c = 6: z +6.59, median d +2.02 pts, certified 100.0%, fallback on 40.6% of plants
36
+ q = 8 c = 7: z +6.72, median d +2.02 pts, certified 100.0%, fallback on 52.1% of plants
37
+ q = 8 c = 8: z +7.70, median d +1.64 pts, certified 100.0%, fallback on 100.0% of plants
38
+ CHOICE m = 2: q = 8, c = 8
results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m2a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c65d064c0ad0a13bde007019679ce814684a48a8c630dc86540450c7e9cca669
3
+ size 44367
results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m2a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78965120819b48c4de3c1b20eb343d9a0a9c13afffc80a56a25e2d07be5049d7
3
+ size 44396
results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m2a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25cbfda8091f8c370164255e275731560edbc94be6dbdd19645e5df68c578fff
3
+ size 44401
results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m3a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e419af63511d92d9c1de2db810d145d107b5a106ccf93e9a2ab1f4fb00abbc2c
3
+ size 99804
results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m3a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b546701925a164e21077e20d1e50b153468a2f767a8c97cd90294c35d4dcf096
3
+ size 99780
results/icml_r3/X28/mimo/gen_fb/ho_fb_x28m3a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9627bde758dd8be1aedd79e1e83910d1d6aaaa3d761fcefc197e39a52d1a3e44
3
+ size 99735
results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m2a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b383625f33d4b079f01207c808f97f30e60271204cff216c726b4d595ef31df
3
+ size 55978
results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m2a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ef4844be63683ee90afaae4eed872d9d32919d5df1f54943d0d1038bf5b1978
3
+ size 55979
results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m2a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7d2d43b43ea96a736c122ab900922c1b307dd000fc283806e035e47f5822913
3
+ size 55990
results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m3a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dd5da7135b856f63651f2976da8a512e44fcce9fac9c0820a613f15ceffcfc5d
3
+ size 111339
results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m3a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9030d2f1358ad2a698cba5461d38f5ba3a5fd8a96c5bb8b80313368128583d67
3
+ size 111397
results/icml_r3/X28/mimo/gen_fb/ho_fbeval_x28m3a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72ed3acd9fefe174115d18497b995fb6e65ac0cac8a3fbfc703b03b09e37a284
3
+ size 110995
results/icml_r3/X28/mimo/gen_fb/t2_c_x28m2c_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da895453a8e184fd8bf11fd74e52047f8e35ad3e3afb523d9e4fa272ab0fd1f9
3
+ size 266295
results/icml_r3/X28/mimo/gen_fb/t2_c_x28m2c_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f93abac56553a60f7cf8a0305d0b5238ed56d056eed60547f6c82457657e9a3
3
+ size 266082
results/icml_r3/X28/mimo/gen_fb/t2_c_x28m2c_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9be3d72d8ed35915cf0e36dfa73e45dd6afdc6f72094c22494182a2bfd5ef563
3
+ size 266041
results/icml_r3/X28/mimo/gen_fb/t2_c_x28m3c_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f719ef04a6b1689378eb88a3877152c3adbfd3285cbaf6ac277af8e10124991
3
+ size 457542
results/icml_r3/X28/mimo/gen_fb/t2_c_x28m3c_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ceda43fe5ea4dfda1618f660b88534379c2c206848511874e927c31c155b1dd6
3
+ size 456717
results/icml_r3/X28/mimo/gen_fb/t2_c_x28m3c_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:63df3005742fb448bee2d27811adca01cc3567b4a51c74ddf01562105d928cb7
3
+ size 457550
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:91570b936715be78148a7ff4ce86588cb0b43b8c490ebace1179928adc8e271d
3
+ size 98024
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9549a28a52b87af33a6456d03beb3c98e65842cc7a481bfefb693ca43cebfe43
3
+ size 98089
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d23de9fea5bcc527d63a96c177cd4069954e62dea9bf9a28efc2efebdc9c40ac
3
+ size 98047
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2b_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:990f4467ef410cf86ebbf69c3406fa4ef1d4f9d1534339f10388ed479a553bc6
3
+ size 245159
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2b_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:281f006722767689877bf205fb76b86b29f9a902b5db12e27ff643724e54e22c
3
+ size 245107
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m2b_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3bf1315997c95550bc247e4b54336d542eb4420421da36fa0538a5a042355b32
3
+ size 245255
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24978b269e533ebf8f1e5e221af8130cce264c60011d1c945f40610c9e94fc1f
3
+ size 221141
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:194be6474f155c8a29a35707bec4052eca37d2285bddbccd2f5879e56634cbec
3
+ size 221120
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bdc399047ff9a2f600ee246c43b8492532e3f2c48d4983d54d848b90f8f91381
3
+ size 221183
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3b_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53fbb303924a42004193878ec2ec79eedf8eb0221024439683a92551d8811c8a
3
+ size 441086
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3b_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a94204cd8a0c71c0861ff5bde5173b87ebf8adedfc43a58377dbb9f625d9e3f8
3
+ size 441243
results/icml_r3/X28/mimo/gen_fb/t2_cand_x28m3b_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:224a1240161e82485c8cdb08360ddac91b847d88b8557c02905272ed733db4c7
3
+ size 441042
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4fbd3fba264480808657c5f5ed34c0440133f23f27bb0d2e1cf6e4b4f890425e
3
+ size 123128
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1480c730b131d324db0a13ac9e4d096c1dc616ab1a37ece788fa3d811d42af6e
3
+ size 123192
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f05b12318ae5925af4b8be14abf38ed14098deeb2411308a9e11dabccafa2def
3
+ size 123102
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2b_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c191a3e3a24ff66b9b2126875ca1fcf20413e296bd711569dfb721f6173fec2
3
+ size 712838
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2b_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42505e3dd38f6117667bc5c6da1ac004d5a89fb511e209e2972e158f013605c4
3
+ size 712889
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m2b_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a4e56f4fce8f5208f83c8d73d260969ae78ce7d91f8dc9d28aba97a87b3aa6f8
3
+ size 713528
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3a_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:959f9e13016acd0c43ea5294f9bffd96549e937c8cec0cee686fa248e1e3e1d1
3
+ size 246088
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3a_s1.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:adff27188d4a2415732162d7d3ac9b36fd55ff4302985e0017a68b045f807e4a
3
+ size 246202
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3a_s2.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:46347c0661eb4e7ac3db7c069d1367f305b302136eb2af7a50a6d4501fd26ccf
3
+ size 245795
results/icml_r3/X28/mimo/gen_fb/t2_eval_x28m3b_s0.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:722e929f7319def9cf7d98dbbebcb8426fa75fcb0ecdff29596d3e4f7148af8a
3
+ size 1111626