Capacity-Shaped Binomial Bottleneck / bench_report.json
Beats tuned baseline
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_config": {
9 "lr": 0.01,
10 "epochs": 30,
11 "weight_decay": 0.0001,
12 "prior": 0.0,
13 "mi": 0.0
14 },
15 "sweep": {
16 "best_cfg": {
17 "lr": 0.01,
18 "epochs": 30,
19 "weight_decay": 0.0001,
20 "prior": 0.0,
21 "mi": 0.0
22 },
23 "sweep": [
24 {
25 "cfg": {
26 "lr": 0.001,
27 "epochs": 30,
28 "weight_decay": 0.0,
29 "prior": 0.0,
30 "mi": 0.0
31 },
32 "mean": 199.0803451538086
33 },
34 {
35 "cfg": {
36 "lr": 0.001,
37 "epochs": 30,
38 "weight_decay": 0.0001,
39 "prior": 0.0,
40 "mi": 0.0
41 },
42 "mean": 199.0547637939453
43 },
44 {
45 "cfg": {
46 "lr": 0.003,
47 "epochs": 30,
48 "weight_decay": 0.0,
49 "prior": 0.0,
50 "mi": 0.0
51 },
52 "mean": 158.36765670776367
53 },
54 {
55 "cfg": {
56 "lr": 0.003,
57 "epochs": 30,
58 "weight_decay": 0.0001,
59 "prior": 0.0,
60 "mi": 0.0
61 },
62 "mean": 157.79228591918945
63 },
64 {
65 "cfg": {
66 "lr": 0.01,
67 "epochs": 30,
68 "weight_decay": 0.0,
69 "prior": 0.0,
70 "mi": 0.0
71 },
72 "mean": 69.26650905609131
73 },
74 {
75 "cfg": {
76 "lr": 0.01,
77 "epochs": 30,
78 "weight_decay": 0.0001,
79 "prior": 0.0,
80 "mi": 0.0
81 },
82 "mean": 66.48912048339844
83 }
84 ],
85 "full": {
86 "mean": 71.74383687973022,
87 "std": 21.879479928715813,
88 "per_seed": [
89 64.52406311035156,
90 65.3621597290039,
91 87.43315887451172,
92 48.63710021972656,
93 122.22001647949219,
94 54.124393463134766,
95 62.05601501464844,
96 69.59378814697266
97 ],
98 "n": 8
99 }
100 },
101 "full": {
102 "mean": 71.74383687973022,
103 "std": 23.390147912586205,
104 "per_seed": [
105 64.52406311035156,
106 65.3621597290039,
107 87.43315887451172,
108 48.63710021972656,
109 122.22001647949219,
110 54.124393463134766,
111 62.05601501464844,
112 69.59378814697266
113 ]
114 }
115 },
116 "idea": {
117 "mean": 70.86578416824341,
118 "std": 23.695512986136727,
119 "per_seed": [
120 64.52386474609375,
121 60.14493942260742,
122 87.41535949707031,
123 48.62728500366211,
124 122.21966552734375,
125 54.03986740112305,
126 62.055946350097656,
127 67.89934539794922
128 ],
129 "config": {
130 "lr": 0.01,
131 "epochs": 30,
132 "weight_decay": 0.0001,
133 "prior": 0.04,
134 "mi": 0.01
135 }
136 },
137 "comparison": {
138 "delta_mean": -0.8780527114868164,
139 "idea_wins": 8,
140 "n_pairs": 8,
141 "per_seed_diffs": [
142 -0.0001983642578125,
143 -5.217220306396484,
144 -0.01779937744140625,
145 -0.009815216064453125,
146 -0.0003509521484375,
147 -0.08452606201171875,
148 -6.866455078125e-05,
149 -1.6944427490234375
150 ],
151 "p_value": 0.0081,
152 "mde": 1.5460051121505387,
153 "mde_rel_pct": 2.1548960571236626,
154 "verdict": "idea better (significant)",
155 "system_worked": true
156 },
157 "audit": {
158 "idea_configs": [
159 {
160 "lr": 0.001,
161 "epochs": 30,
162 "weight_decay": 0.0001,
163 "prior": 0.04,
164 "mi": 0.01
165 },
166 {
167 "lr": 0.003,
168 "epochs": 30,
169 "weight_decay": 0.0001,
170 "prior": 0.04,
171 "mi": 0.01
172 },
173 {
174 "lr": 0.01,
175 "epochs": 30,
176 "weight_decay": 0.0001,
177 "prior": 0.04,
178 "mi": 0.01
179 }
180 ],
181 "baseline_lrs": [
182 0.001,
183 0.003,
184 0.01
185 ],
186 "device": "cuda"
187 },
188 "mechanism_signature": {
189 "predicted_prior_endpoint_mass": 0.39276123046875067,
190 "observed_endpoint_mass": 0.9991894885897636,
191 "predicted_prior_l1_to_reference": 0.0,
192 "observed_prior_l1_to_reference": 1.2845361977815628,
193 "observed_test_MI_nats": 0.00037778520345455036,
194 "observed_test_KL": 1.6237996518611908,
195 "confirmed": false
196 }
197}