Эксперимент: TD-to-PDE Continuation Training (#943)

{ "worked": true, "confidence": 8, "verdict": "Built a 1-D OU diffusion critic experiment with automatic-differentiation generator residuals, TD loss, boundary penalty, fixed-PDE, TD-only, and gated continuation schedules. The generator check passed exactly, and weak-Euler error decreased from 0.00796 to 0.000922 as dt fell by 8x, close to the predicted linear ratio 0.125 (observed 0.116). The continuation gate showed a threshold transition, and activated continuation achieved lower PDE loss than fixed weighting, although fixed weighting had slightly better value RMSE and TD loss; this is a promising mechanism signal rather than a decisive overall win.", "metrics": { "baseline": "TD-only: TD loss 0.000911, PDE loss 0.537, value RMSE 0.375. Fixed PDE weight 10: TD loss 0.004876, PDE loss 0.001084, value RMSE 0.02934.", "idea": "Continuation at TD threshold 0.002: first ramp at step 5, final lambda 10, TD loss 0.005029, PDE loss 0.0000661, value RMSE 0.03167. Generator autodiff error 0; weak-generator error ratio 0.116 observed versus 0.125 predicted." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 run_experiment.py", "files": [ "run_experiment.py", "results.json" ], "limitations": "Only a small manufactured 1-D diffusion and one random seed were tested; no actor learning, action maximization, reach-avoid boundaries, stochastic-policy evaluation, hyperparameter robustness, wall-clock comparison, or larger architectures were evaluated." }