Doubled-angle orientation order pooling / bench_report.json
Failed on benchmark
1{
2 "bench_version": 1,
3 "track": "oriented_line_quotient",
4 "model": "cnn_small",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.003,
10 "weight_decay": 0.0
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.1999999973922968
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "weight_decay": 0.0001
24 },
25 "mean": 0.1287499964237213
26 },
27 {
28 "cfg": {
29 "lr": 0.003,
30 "weight_decay": 0.0
31 },
32 "mean": 0.06999999936670065
33 },
34 {
35 "cfg": {
36 "lr": 0.003,
37 "weight_decay": 0.0001
38 },
39 "mean": 0.13249999564141035
40 },
41 {
42 "cfg": {
43 "lr": 0.006,
44 "weight_decay": 0.0
45 },
46 "mean": 0.2887499984353781
47 },
48 {
49 "cfg": {
50 "lr": 0.006,
51 "weight_decay": 0.0001
52 },
53 "mean": 0.358749995008111
54 }
55 ],
56 "full": {
57 "mean": 0.16249999217689037,
58 "std": 0.17727096625015212,
59 "per_seed": [
60 0.07000000029802322,
61 0.1599999964237213,
62 0.06499999761581421,
63 0.03500000014901161,
64 0.2149999886751175,
65 0.11499999463558197,
66 0.6049999594688416,
67 0.03500000014901161
68 ],
69 "n": 8
70 },
71 "architecture_note": "fixed oriented Sobel bank + scalar mean + shared CNN"
72 },
73 "idea": {
74 "cfg": {
75 "lr": 0.003,
76 "weight_decay": 0.0
77 },
78 "mean": 0.10562499891966581,
79 "std": 0.047265044108773906,
80 "per_seed": [
81 0.08999999612569809,
82 0.07999999821186066,
83 0.1550000011920929,
84 0.07999999821186066,
85 0.1899999976158142,
86 0.07000000029802322,
87 0.14000000059604645,
88 0.03999999910593033
89 ],
90 "n": 8,
91 "architecture_note": "fixed oriented Sobel bank + doubled-angle zR,zI,q + shared CNN",
92 "trials": [
93 {
94 "cfg": {
95 "lr": 0.001,
96 "weight_decay": 0.0
97 },
98 "mean": 0.10750000039115548,
99 "std": 0.057445627674003165,
100 "per_seed": [
101 0.04999999701976776,
102 0.03500000014901161,
103 0.17000000178813934,
104 0.08500000089406967,
105 0.03999999910593033,
106 0.17000000178813934,
107 0.17000000178813934,
108 0.14000000059604645
109 ],
110 "n": 8
111 },
112 {
113 "cfg": {
114 "lr": 0.001,
115 "weight_decay": 0.0001
116 },
117 "mean": 0.10999999684281647,
118 "std": 0.06689543906243255,
119 "per_seed": [
120 0.03500000014901161,
121 0.02499999850988388,
122 0.1899999976158142,
123 0.11499999463558197,
124 0.029999999329447746,
125 0.1599999964237213,
126 0.19499999284744263,
127 0.12999999523162842
128 ],
129 "n": 8
130 },
131 {
132 "cfg": {
133 "lr": 0.003,
134 "weight_decay": 0.0
135 },
136 "mean": 0.10562499891966581,
137 "std": 0.047265044108773906,
138 "per_seed": [
139 0.08999999612569809,
140 0.07999999821186066,
141 0.1550000011920929,
142 0.07999999821186066,
143 0.1899999976158142,
144 0.07000000029802322,
145 0.14000000059604645,
146 0.03999999910593033
147 ],
148 "n": 8
149 },
150 {
151 "cfg": {
152 "lr": 0.003,
153 "weight_decay": 0.0001
154 },
155 "mean": 0.12499999906867743,
156 "std": 0.02936835116105255,
157 "per_seed": [
158 0.1550000011920929,
159 0.1550000011920929,
160 0.09999999403953552,
161 0.08500000089406967,
162 0.07999999821186066,
163 0.14000000059604645,
164 0.14000000059604645,
165 0.14499999582767487
166 ],
167 "n": 8
168 },
169 {
170 "cfg": {
171 "lr": 0.006,
172 "weight_decay": 0.0
173 },
174 "mean": 0.1518749939277768,
175 "std": 0.08975443257541645,
176 "per_seed": [
177 0.09999999403953552,
178 0.125,
179 0.3349999785423279,
180 0.07000000029802322,
181 0.14000000059604645,
182 0.07999999821186066,
183 0.09999999403953552,
184 0.26499998569488525
185 ],
186 "n": 8
187 },
188 {
189 "cfg": {
190 "lr": 0.006,
191 "weight_decay": 0.0001
192 },
193 "mean": 0.23749999701976776,
194 "std": 0.10037429990025486,
195 "per_seed": [
196 0.22499999403953552,
197 0.3449999988079071,
198 0.3050000071525574,
199 0.125,
200 0.32499998807907104,
201 0.07999999821186066,
202 0.3449999988079071,
203 0.14999999105930328
204 ],
205 "n": 8
206 }
207 ]
208 },
209 "comparison": {
210 "delta_mean": -0.05687499325722456,
211 "idea_wins": 4,
212 "n_pairs": 8,
213 "per_seed_diffs": [
214 0.019999995827674866,
215 -0.07999999821186066,
216 0.09000000357627869,
217 0.044999998062849045,
218 -0.024999991059303284,
219 -0.044999994337558746,
220 -0.4649999588727951,
221 0.004999998956918716
222 ],
223 "p_value": 0.54375,
224 "mde": 0.14480774970584148,
225 "mde_rel_pct": 89.11246564751222,
226 "verdict": "no significant win",
227 "system_worked": false
228 },
229 "mechanism_signature": {
230 "predicted": {
231 "pi_rotation_mean_pool_error": 0.0,
232 "q_pi_rotation_error": 0.0,
233 "90deg_doubled_angle_relation": "zR,zI should sign-flip"
234 },
235 "observed": {
236 "pi_rotation_mean_pool_error": 0.09535165876150131,
237 "q_pi_rotation_error": 0.007520223036408424,
238 "90deg_relation_abs_error": 0.23041518032550812
239 },
240 "tolerance": {
241 "pi_rotation_mean_pool_error": 0.03,
242 "q_pi_rotation_error": 0.03,
243 "90deg_relation_abs_error": 0.1
244 },
245 "confirmed": false,
246 "measurement": "trained idea model on held-out CIFAR test images"
247 },
248 "custom_track": {
249 "name": "oriented_line_quotient",
250 "file": "custom_orientation_lines.py",
251 "domain": "vision_orientation"
252 },
253 "protocol_note": "custom oriented-line subset, 400/200 samples, 8 epochs, 8 paired seeds; baseline sweep uses 4 seeds then full best config."
254}