device=cuda /home/cercola/matte_server/DiffusionControl/nonlinear/data_generation/pendulum_lmi.py:149: UserWarning: Solution may be inaccurate. Try another solver, adjusting the solver settings, or solve with verbose=True for more information. prob.solve(solver=solver, verbose=False) [gen-baseline/pendulum] draw seed=0 done [gen-baseline/pendulum] draw seed=1 done [gen-baseline/pendulum] draw seed=2 done === A3 GENERATIVE BASELINE — PENDULUM (N=40, shared cached oracle, 3 draws/K, seeds=[0, 1, 2]) === budget: diffusion params=3,823,883 epochs=800 time_to_best=232.0s; cvae params=277,019 epochs=800 time_to_best=278.9s -- K=1 -- diffusion : stable>=1 1.000+-0.000 cert>=1 0.992+-0.014 persample 1.000+-0.000 median_regret +1.52%+-0.06pp diversity 0.0000+-0.0000 cvae : stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.37%+-0.05pp diversity 0.0000+-0.0000 paired McNemar (diffusion vs cvae, pooled 120 plant x draw trials): cvae wins 0/0 discordant, p=1 -- K=8 -- diffusion : stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.11%+-0.09pp diversity 0.0142+-0.0007 cvae : stable>=1 1.000+-0.000 cert>=1 1.000+-0.000 persample 1.000+-0.000 median_regret +1.20%+-0.02pp diversity 0.0027+-0.0001 paired McNemar (diffusion vs cvae, pooled 120 plant x draw trials): cvae wins 0/0 discordant, p=1 [timing] pendulum arm: 131.6s [gen-baseline/siso] draw seed=0 done [gen-baseline/siso] draw seed=1 done [gen-baseline/siso] draw seed=2 done === A3 GENERATIVE BASELINE — SISO (N=60 plants, plant_seed=100, 3 draws/K, seeds=[0, 1, 2]) === budget: diffusion params=21,445,653 epochs=1200 time_to_best=7940.6s; cvae params=1,108,021 epochs=60 time_to_best=33.9s; regressor params=804,373 epochs=40 time=not recorded -- K=1 -- diffusion : plant_stabilisable 0.667+-0.033 persample 0.667+-0.033 median_best_J 6.713+-3.633 diversity 0.0000+-0.0000 cvae : plant_stabilisable 0.750+-0.050 persample 0.750+-0.050 median_best_J 5.167+-2.512 diversity 0.0000+-0.0000 regressor : plant_stabilisable 0.750+-0.044 persample 0.750+-0.044 median_best_J 2.403+-0.141 diversity 0.0000+-0.0000 paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 18/21 discordant, p=0.00149 paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 19/23 discordant, p=0.0026 -- K=8 -- diffusion : plant_stabilisable 0.800+-0.050 persample 0.640+-0.024 median_best_J 1.363+-0.120 diversity 0.1844+-0.0076 cvae : plant_stabilisable 0.767+-0.033 persample 0.751+-0.045 median_best_J 0.748+-0.113 diversity 0.0034+-0.0001 regressor : plant_stabilisable 0.750+-0.044 persample 0.750+-0.044 median_best_J 2.403+-0.141 diversity 0.0000+-0.0000 paired McNemar (diffusion vs cvae, pooled 180 plant x draw trials): cvae wins 0/6 discordant, p=0.0312 paired McNemar (diffusion vs regressor, pooled 180 plant x draw trials): regressor wins 0/9 discordant, p=0.00391 [timing] siso arm: 488.6s