Эксперимент: Uncertainty-Propagation Tree Acquisition (#438)

{ "worked": false, "confidence": 9, "verdict": "Built an exact RBF-GP uncertainty-propagation acquisition prototype with depth-2 exhaustive tree planning, propagated hypothetical posterior variance, transition costs, parameter sweeps, and a random/greedy comparison. Variance non-increase was confirmed (maximum numerical increase -3.05e-9), and the cost-policy crossover matched the analytic prediction 0.55103 versus observed grid midpoint 0.55150. However, the redundancy sweep was not monotonic and the tree matched greedy exactly (cumulative reduction 0.7204 for both), so the proposed acquisition win was not demonstrated.", "metrics": { "baseline": "Greedy: cumulative variance reduction 0.7204116062; final mean variance 0.1241876201; actions [2,4,0]. Random: reduction 0.4426903473; final mean variance 0.4019088791.", "idea": "Tree: cumulative variance reduction 0.7204116062; final mean variance 0.1241876201; actions [2,4,0]. Variance stability max increase -3.0538e-9; cost crossover predicted lambda 0.5510273 versus observed 0.5515; redundancy monotonicity failed, though endpoint reward ratio was 0.0763." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 verify.py", "files": [ "experiment.py", "verify.py", "results.json" ], "limitations": "This is a small one-dimensional exact GP toy, not CIFAR-10 or a neural ensemble. It uses exhaustive depth-2 search rather than Monte Carlo rollouts/UCB, has synthetic spatial batches and costs, tests only three acquisition steps, and does not measure labeling time, FLOPs, calibration, predictive accuracy, or scalability." }