Correction-aware tree optimizer / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "tabular",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.01,
10 "rho": 0.0,
11 "penalty": 0.0,
12 "epochs": 20
13 },
14 "sweep": [
15 {
16 "cfg": {
17 "lr": 0.001,
18 "rho": 0.0,
19 "penalty": 0.0,
20 "epochs": 20
21 },
22 "mean": 215.17152404785156
23 },
24 {
25 "cfg": {
26 "lr": 0.001,
27 "rho": 0.05,
28 "penalty": 0.1,
29 "epochs": 20
30 },
31 "mean": 215.17152404785156
32 },
33 {
34 "cfg": {
35 "lr": 0.001,
36 "rho": 0.2,
37 "penalty": 0.8,
38 "epochs": 20
39 },
40 "mean": 215.17152404785156
41 },
42 {
43 "cfg": {
44 "lr": 0.003,
45 "rho": 0.0,
46 "penalty": 0.0,
47 "epochs": 20
48 },
49 "mean": 72.50996971130371
50 },
51 {
52 "cfg": {
53 "lr": 0.003,
54 "rho": 0.05,
55 "penalty": 0.1,
56 "epochs": 20
57 },
58 "mean": 72.50996971130371
59 },
60 {
61 "cfg": {
62 "lr": 0.003,
63 "rho": 0.2,
64 "penalty": 0.8,
65 "epochs": 20
66 },
67 "mean": 72.50996971130371
68 },
69 {
70 "cfg": {
71 "lr": 0.01,
72 "rho": 0.0,
73 "penalty": 0.0,
74 "epochs": 20
75 },
76 "mean": 24.599658966064453
77 },
78 {
79 "cfg": {
80 "lr": 0.01,
81 "rho": 0.05,
82 "penalty": 0.1,
83 "epochs": 20
84 },
85 "mean": 24.599658966064453
86 },
87 {
88 "cfg": {
89 "lr": 0.01,
90 "rho": 0.2,
91 "penalty": 0.8,
92 "epochs": 20
93 },
94 "mean": 24.599658966064453
95 }
96 ],
97 "full": {
98 "mean": 31.372122287750244,
99 "std": 9.015775011632964,
100 "per_seed": [
101 22.099992752075195,
102 30.866003036499023,
103 23.188148498535156,
104 22.244491577148438,
105 28.95732879638672,
106 48.877689361572266,
107 33.53181076049805,
108 41.21151351928711
109 ],
110 "n": 8
111 },
112 "method": "FedAvg (central mean after local minibatch step)"
113 },
114 "idea": {
115 "mean": 42.09156918525696,
116 "std": 19.275065768488197,
117 "per_seed": [
118 72.97396087646484,
119 40.06073760986328,
120 67.2032470703125,
121 53.84383010864258,
122 21.086515426635742,
123 30.523590087890625,
124 33.312156677246094,
125 17.728515625
126 ],
127 "n": 8
128 },
129 "comparison": {
130 "delta_mean": 10.719446897506714,
131 "idea_wins": 4,
132 "n_pairs": 8,
133 "per_seed_diffs": [
134 50.87396812438965,
135 9.194734573364258,
136 44.015098571777344,
137 31.59933853149414,
138 -7.870813369750977,
139 -18.35409927368164,
140 -0.21965408325195312,
141 -23.48299789428711
142 ],
143 "p_value": 0.2999,
144 "mde": 23.733515768986862,
145 "mde_rel_pct": 75.65161053274996,
146 "verdict": "no significant win",
147 "system_worked": false
148 },
149 "mechanism_signature": {
150 "mechanism": "trained four-worker tree residual and dual correction measurements",
151 "predicted": "redistributed corrections should be nonzero and reduce stale-consensus error",
152 "observed": [
153 {
154 "seed": 0,
155 "final_residual": 7.351737976074219,
156 "correction_norm": 0.4109642207622528,
157 "initial_residual": 9.530355453491211
158 },
159 {
160 "seed": 1,
161 "final_residual": 7.59358024597168,
162 "correction_norm": 0.3979034423828125,
163 "initial_residual": 9.610031127929688
164 },
165 {
166 "seed": 2,
167 "final_residual": 7.465011119842529,
168 "correction_norm": 0.3986107110977173,
169 "initial_residual": 9.460624694824219
170 },
171 {
172 "seed": 3,
173 "final_residual": 7.670905590057373,
174 "correction_norm": 0.40616631507873535,
175 "initial_residual": 9.607115745544434
176 },
177 {
178 "seed": 4,
179 "final_residual": 7.9027018547058105,
180 "correction_norm": 0.37859806418418884,
181 "initial_residual": 9.442831039428711
182 },
183 {
184 "seed": 5,
185 "final_residual": 7.862105369567871,
186 "correction_norm": 0.386971652507782,
187 "initial_residual": 9.534515380859375
188 },
189 {
190 "seed": 6,
191 "final_residual": 7.549962043762207,
192 "correction_norm": 0.40791869163513184,
193 "initial_residual": 9.64505386352539
194 },
195 {
196 "seed": 7,
197 "final_residual": 7.712738037109375,
198 "correction_norm": 0.4060078263282776,
199 "initial_residual": 9.527835845947266
200 }
201 ],
202 "confirmed": true
203 },
204 "idea_sweep": [
205 {
206 "cfg": {
207 "lr": 0.01,
208 "rho": 0.0,
209 "penalty": 0.0,
210 "epochs": 20
211 },
212 "result": {
213 "mean": 47.336907148361206,
214 "std": 28.887751826226218,
215 "per_seed": [
216 22.092214584350586,
217 35.96384048461914,
218 73.7575454711914,
219 22.593856811523438,
220 88.00531005859375,
221 27.3686466217041,
222 89.42167663574219,
223 19.49216651916504
224 ],
225 "n": 8
226 }
227 },
228 {
229 "cfg": {
230 "lr": 0.01,
231 "rho": 0.05,
232 "penalty": 0.1,
233 "epochs": 20
234 },
235 "result": {
236 "mean": 42.09156918525696,
237 "std": 19.275065768488197,
238 "per_seed": [
239 72.97396087646484,
240 40.06073760986328,
241 67.2032470703125,
242 53.84383010864258,
243 21.086515426635742,
244 30.523590087890625,
245 33.312156677246094,
246 17.728515625
247 ],
248 "n": 8
249 }
250 },
251 {
252 "cfg": {
253 "lr": 0.01,
254 "rho": 0.2,
255 "penalty": 0.8,
256 "epochs": 20
257 },
258 "result": {
259 "mean": 43.77399754524231,
260 "std": 15.561092309534422,
261 "per_seed": [
262 28.526273727416992,
263 48.15719223022461,
264 52.14424514770508,
265 37.76236343383789,
266 71.31439208984375,
267 55.57741928100586,
268 18.206344604492188,
269 38.50374984741211
270 ],
271 "n": 8
272 }
273 }
274 ],
275 "selection_note": "Baseline and idea use the same lr/rho/penalty union; baseline ignores tree knobs as standard FedAvg."
276}