AbstractPhil commited on
Commit
8c10e31
·
verified ·
1 Parent(s): d4f5cbf

Upload weights - GeoFractalDavid-Basin-k4 - Run 20251015_230643 - Acc 80.51%

Browse files
weights/GeoFractalDavid-Basin-k4/20251015_230643/README.md CHANGED
@@ -23,7 +23,7 @@ model-index:
23
  type: imagenet-1k
24
  metrics:
25
  - type: accuracy
26
- value: 79.00
27
  name: Validation Accuracy
28
  ---
29
 
@@ -34,18 +34,18 @@ Features must "fit" geometric signatures: k-simplex shapes, Cantor positions, an
34
 
35
  ## 🎯 Performance
36
 
37
- - **Best Validation Accuracy**: 79.00%
38
- - **Epoch**: 2/10
39
- - **Training Time**: 5m
40
 
41
  ### Per-Scale Performance
42
- - **Scale 576D**: 78.33%
43
- - **Scale 640D**: 78.40%
44
- - **Scale 704D**: 78.29%
45
- - **Scale 768D**: 78.44%
46
- - **Scale 832D**: 78.29%
47
- - **Scale 896D**: 77.50%
48
- - **Scale 960D**: 76.88%
49
 
50
 
51
  ## 🏗️ Architecture
@@ -73,8 +73,8 @@ Each scale learns to weight these components differently.
73
  The alpha parameter controls middle-interval weighting in the Cantor staircase.
74
 
75
  - **Initial**: 0.4817
76
- - **Final**: 0.4149
77
- - **Change**: -0.0668
78
  - **Converged to 0.5**: False
79
 
80
  The Cantor staircase uses soft triadic decomposition with learnable alpha to map
@@ -86,39 +86,39 @@ Each class has a learned scalar Cantor prototype. The model pulls features towar
86
  their class's Cantor position.
87
 
88
  **Scale 576D**:
89
- - Mean: 0.4246
90
- - Std: 0.1212
91
- - Range: [0.0008, 0.5805]
92
 
93
  **Scale 640D**:
94
- - Mean: 0.4246
95
- - Std: 0.1212
96
- - Range: [0.0011, 0.5805]
97
 
98
  **Scale 704D**:
99
- - Mean: 0.4256
100
- - Std: 0.1205
101
- - Range: [0.0018, 0.5809]
102
 
103
  **Scale 768D**:
104
- - Mean: 0.4255
105
- - Std: 0.1118
106
- - Range: [0.0306, 0.5832]
107
 
108
  **Scale 832D**:
109
- - Mean: 0.4254
110
- - Std: 0.1205
111
- - Range: [0.0013, 0.5810]
112
 
113
  **Scale 896D**:
114
- - Mean: 0.4243
115
- - Std: 0.1170
116
- - Range: [0.0052, 0.5829]
117
 
118
  **Scale 960D**:
119
- - Mean: 0.4252
120
- - Std: 0.1174
121
- - Range: [0.0020, 0.5841]
122
 
123
 
124
  Most classes cluster around 0.5 (middle Cantor region), with smooth spread across [0,1].
@@ -128,13 +128,13 @@ This creates a continuous manifold rather than discrete bins.
128
 
129
  Each scale learns optimal weights for combining geometric components:
130
 
131
- **Scale 576D**: Feature=0.681, Cantor=0.073, Crystal=0.246
132
- **Scale 640D**: Feature=0.659, Cantor=0.075, Crystal=0.266
133
- **Scale 704D**: Feature=0.625, Cantor=0.073, Crystal=0.302
134
- **Scale 768D**: Feature=0.849, Cantor=0.028, Crystal=0.123
135
- **Scale 832D**: Feature=0.632, Cantor=0.061, Crystal=0.307
136
- **Scale 896D**: Feature=0.742, Cantor=0.038, Crystal=0.220
137
- **Scale 960D**: Feature=0.688, Cantor=0.041, Crystal=0.271
138
 
139
 
140
  **Pattern**: Lower scales rely on feature similarity, higher scales use crystal geometry.
 
23
  type: imagenet-1k
24
  metrics:
25
  - type: accuracy
26
+ value: 80.51
27
  name: Validation Accuracy
28
  ---
29
 
 
34
 
35
  ## 🎯 Performance
36
 
37
+ - **Best Validation Accuracy**: 80.51%
38
+ - **Epoch**: 4/10
39
+ - **Training Time**: 10m
40
 
41
  ### Per-Scale Performance
42
+ - **Scale 576D**: 76.82%
43
+ - **Scale 640D**: 76.96%
44
+ - **Scale 704D**: 76.89%
45
+ - **Scale 768D**: 79.47%
46
+ - **Scale 832D**: 76.11%
47
+ - **Scale 896D**: 70.58%
48
+ - **Scale 960D**: 68.24%
49
 
50
 
51
  ## 🏗️ Architecture
 
73
  The alpha parameter controls middle-interval weighting in the Cantor staircase.
74
 
75
  - **Initial**: 0.4817
76
+ - **Final**: 0.3009
77
+ - **Change**: -0.1808
78
  - **Converged to 0.5**: False
79
 
80
  The Cantor staircase uses soft triadic decomposition with learnable alpha to map
 
86
  their class's Cantor position.
87
 
88
  **Scale 576D**:
89
+ - Mean: 0.3247
90
+ - Std: 0.1574
91
+ - Range: [-0.0008, 0.5011]
92
 
93
  **Scale 640D**:
94
+ - Mean: 0.3242
95
+ - Std: 0.1579
96
+ - Range: [-0.0004, 0.5008]
97
 
98
  **Scale 704D**:
99
+ - Mean: 0.3246
100
+ - Std: 0.1577
101
+ - Range: [-0.0007, 0.5009]
102
 
103
  **Scale 768D**:
104
+ - Mean: 0.3301
105
+ - Std: 0.1542
106
+ - Range: [-0.0007, 0.5011]
107
 
108
  **Scale 832D**:
109
+ - Mean: 0.3249
110
+ - Std: 0.1574
111
+ - Range: [-0.0006, 0.5012]
112
 
113
  **Scale 896D**:
114
+ - Mean: 0.3270
115
+ - Std: 0.1561
116
+ - Range: [-0.0007, 0.5013]
117
 
118
  **Scale 960D**:
119
+ - Mean: 0.3274
120
+ - Std: 0.1560
121
+ - Range: [-0.0004, 0.5011]
122
 
123
 
124
  Most classes cluster around 0.5 (middle Cantor region), with smooth spread across [0,1].
 
128
 
129
  Each scale learns optimal weights for combining geometric components:
130
 
131
+ **Scale 576D**: Feature=0.729, Cantor=0.046, Crystal=0.225
132
+ **Scale 640D**: Feature=0.695, Cantor=0.048, Crystal=0.257
133
+ **Scale 704D**: Feature=0.637, Cantor=0.049, Crystal=0.314
134
+ **Scale 768D**: Feature=0.957, Cantor=0.005, Crystal=0.037
135
+ **Scale 832D**: Feature=0.634, Cantor=0.041, Crystal=0.325
136
+ **Scale 896D**: Feature=0.794, Cantor=0.011, Crystal=0.195
137
+ **Scale 960D**: Feature=0.643, Cantor=0.017, Crystal=0.340
138
 
139
 
140
  **Pattern**: Lower scales rely on feature similarity, higher scales use crystal geometry.
weights/GeoFractalDavid-Basin-k4/20251015_230643/model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7c946e4015c9904c593a66d0991622acba8e2c70527ac16533ab73f9d7e11711
3
  size 174686092
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4245eb6954d05b5b524dcde0e7b2af05bcf65c85d9ba7b11a83294cf4b141456
3
  size 174686092
weights/GeoFractalDavid-Basin-k4/20251015_230643/model_metadata.json CHANGED
@@ -1,21 +1,21 @@
1
  {
2
- "epoch": 1,
3
  "metrics": {
4
- "val_acc": 79.002,
5
- "train_acc": 80.24426167704912,
6
  "scale_accuracies": {
7
- "576": 78.328,
8
- "640": 78.396,
9
- "704": 78.294,
10
- "768": 78.442,
11
- "832": 78.292,
12
- "896": 77.498,
13
- "960": 76.88
14
  },
15
- "best_val_acc": 79.002,
16
- "best_epoch": 1,
17
- "final_train_acc": 80.24426167704912,
18
- "training_time": "5m"
19
  },
20
  "config": {
21
  "name": "geofractal_david_basin",
@@ -72,126 +72,136 @@
72
  "alpha_summary": {
73
  "global": {
74
  "initial": 0.4817016124725342,
75
- "final": 0.414915531873703,
76
- "change": -0.06678608059883118,
77
  "converged_to_0.5": false
78
  }
79
  },
80
  "cantor_prototypes": {
81
  "576": {
82
- "final_mean": 0.42456430196762085,
83
- "final_std": 0.12118027359247208,
84
  "final_range": [
85
- 0.0008374242461286485,
86
- 0.5804639458656311
87
  ]
88
  },
89
  "640": {
90
- "final_mean": 0.42460644245147705,
91
- "final_std": 0.12123845517635345,
92
  "final_range": [
93
- 0.0011197840794920921,
94
- 0.5804531574249268
95
  ]
96
  },
97
  "704": {
98
- "final_mean": 0.4256438910961151,
99
- "final_std": 0.12051156163215637,
100
  "final_range": [
101
- 0.0018036324763670564,
102
- 0.5808702111244202
103
  ]
104
  },
105
  "768": {
106
- "final_mean": 0.42545634508132935,
107
- "final_std": 0.11182307451963425,
108
  "final_range": [
109
- 0.030573837459087372,
110
- 0.5831647515296936
111
  ]
112
  },
113
  "832": {
114
- "final_mean": 0.4253726005554199,
115
- "final_std": 0.12047051638364792,
116
  "final_range": [
117
- 0.001257226336747408,
118
- 0.5809960961341858
119
  ]
120
  },
121
  "896": {
122
- "final_mean": 0.42428281903266907,
123
- "final_std": 0.11696494370698929,
124
  "final_range": [
125
- 0.005223999731242657,
126
- 0.5829265117645264
127
  ]
128
  },
129
  "960": {
130
- "final_mean": 0.42520272731781006,
131
- "final_std": 0.11736594140529633,
132
  "final_range": [
133
- 0.0020108737517148256,
134
- 0.5841002464294434
135
  ]
136
  }
137
  },
138
  "geo_weights": {
139
  "576": {
140
- "feature": 0.6808567643165588,
141
- "cantor": 0.07346183806657791,
142
- "crystal": 0.24568137526512146
143
  },
144
  "640": {
145
- "feature": 0.6590990424156189,
146
- "cantor": 0.07502132654190063,
147
- "crystal": 0.26587963104248047
148
  },
149
  "704": {
150
- "feature": 0.6252120137214661,
151
- "cantor": 0.07320406287908554,
152
- "crystal": 0.3015839457511902
153
  },
154
  "768": {
155
- "feature": 0.8492695093154907,
156
- "cantor": 0.028041880577802658,
157
- "crystal": 0.12268858402967453
158
  },
159
  "832": {
160
- "feature": 0.6320995092391968,
161
- "cantor": 0.06110195070505142,
162
- "crystal": 0.3067985475063324
163
  },
164
  "896": {
165
- "feature": 0.7415841221809387,
166
- "cantor": 0.03822467848658562,
167
- "crystal": 0.22019124031066895
168
  },
169
  "960": {
170
- "feature": 0.6880423426628113,
171
- "cantor": 0.041151370853185654,
172
- "crystal": 0.27080631256103516
173
  }
174
  },
175
  "training_history": {
176
  "epochs": [
177
  1,
178
- 2
 
 
179
  ],
180
  "train_loss": [
181
  1.9880030441779297,
182
- 1.4796098911533722
 
 
183
  ],
184
  "train_acc": [
185
  74.1779955306373,
186
- 80.24426167704912
 
 
187
  ],
188
  "val_acc": [
189
  78.128,
190
- 79.002
 
 
191
  ],
192
  "lr": [
193
  0.001,
194
- 0.0009755527298894294
 
 
195
  ]
196
  }
197
  }
 
1
  {
2
+ "epoch": 3,
3
  "metrics": {
4
+ "val_acc": 80.514,
5
+ "train_acc": 81.97502745543711,
6
  "scale_accuracies": {
7
+ "576": 76.822,
8
+ "640": 76.962,
9
+ "704": 76.888,
10
+ "768": 79.468,
11
+ "832": 76.108,
12
+ "896": 70.584,
13
+ "960": 68.244
14
  },
15
+ "best_val_acc": 80.514,
16
+ "best_epoch": 3,
17
+ "final_train_acc": 81.97502745543711,
18
+ "training_time": "10m"
19
  },
20
  "config": {
21
  "name": "geofractal_david_basin",
 
72
  "alpha_summary": {
73
  "global": {
74
  "initial": 0.4817016124725342,
75
+ "final": 0.3008934259414673,
76
+ "change": -0.1808081865310669,
77
  "converged_to_0.5": false
78
  }
79
  },
80
  "cantor_prototypes": {
81
  "576": {
82
+ "final_mean": 0.32474276423454285,
83
+ "final_std": 0.15744858980178833,
84
  "final_range": [
85
+ -0.0008127436740323901,
86
+ 0.5011425614356995
87
  ]
88
  },
89
  "640": {
90
+ "final_mean": 0.32419049739837646,
91
+ "final_std": 0.1579282432794571,
92
  "final_range": [
93
+ -0.0004192948108538985,
94
+ 0.5007973909378052
95
  ]
96
  },
97
  "704": {
98
+ "final_mean": 0.3245547115802765,
99
+ "final_std": 0.15768137574195862,
100
  "final_range": [
101
+ -0.0006738750380463898,
102
+ 0.5008888244628906
103
  ]
104
  },
105
  "768": {
106
+ "final_mean": 0.33013302087783813,
107
+ "final_std": 0.15423697233200073,
108
  "final_range": [
109
+ -0.0006688383873552084,
110
+ 0.5011330246925354
111
  ]
112
  },
113
  "832": {
114
+ "final_mean": 0.32486245036125183,
115
+ "final_std": 0.157425194978714,
116
  "final_range": [
117
+ -0.0005794782773591578,
118
+ 0.5011799335479736
119
  ]
120
  },
121
  "896": {
122
+ "final_mean": 0.32698965072631836,
123
+ "final_std": 0.15607324242591858,
124
  "final_range": [
125
+ -0.0006874061073176563,
126
+ 0.5012513995170593
127
  ]
128
  },
129
  "960": {
130
+ "final_mean": 0.32736802101135254,
131
+ "final_std": 0.15600250661373138,
132
  "final_range": [
133
+ -0.00039892634958960116,
134
+ 0.5011450052261353
135
  ]
136
  }
137
  },
138
  "geo_weights": {
139
  "576": {
140
+ "feature": 0.7286769151687622,
141
+ "cantor": 0.046016860753297806,
142
+ "crystal": 0.22530628740787506
143
  },
144
  "640": {
145
+ "feature": 0.6948461532592773,
146
+ "cantor": 0.048131298273801804,
147
+ "crystal": 0.2570226192474365
148
  },
149
  "704": {
150
+ "feature": 0.6369317770004272,
151
+ "cantor": 0.04935236647725105,
152
+ "crystal": 0.3137158751487732
153
  },
154
  "768": {
155
+ "feature": 0.957398533821106,
156
+ "cantor": 0.005391538608819246,
157
+ "crystal": 0.03720993548631668
158
  },
159
  "832": {
160
+ "feature": 0.6340004801750183,
161
+ "cantor": 0.04060900956392288,
162
+ "crystal": 0.3253905475139618
163
  },
164
  "896": {
165
+ "feature": 0.7941204309463501,
166
+ "cantor": 0.010744167491793633,
167
+ "crystal": 0.19513538479804993
168
  },
169
  "960": {
170
+ "feature": 0.6430188417434692,
171
+ "cantor": 0.01670054718852043,
172
+ "crystal": 0.34028053283691406
173
  }
174
  },
175
  "training_history": {
176
  "epochs": [
177
  1,
178
+ 2,
179
+ 3,
180
+ 4
181
  ],
182
  "train_loss": [
183
  1.9880030441779297,
184
+ 1.4796098911533722,
185
+ 1.3439062684298324,
186
+ 1.2507981145724703
187
  ],
188
  "train_acc": [
189
  74.1779955306373,
190
+ 80.24426167704912,
191
+ 81.11752800376532,
192
+ 81.97502745543711
193
  ],
194
  "val_acc": [
195
  78.128,
196
+ 79.002,
197
+ 79.918,
198
+ 80.514
199
  ],
200
  "lr": [
201
  0.001,
202
+ 0.0009755527298894294,
203
+ 0.0009046039886902864,
204
+ 0.0007940987335200905
205
  ]
206
  }
207
  }
weights/GeoFractalDavid-Basin-k4/20251015_230643/training_history.json CHANGED
@@ -2,91 +2,133 @@
2
  "training_history": {
3
  "epochs": [
4
  1,
5
- 2
 
 
6
  ],
7
  "train_loss": [
8
  1.9880030441779297,
9
- 1.4796098911533722
 
 
10
  ],
11
  "train_acc": [
12
  74.1779955306373,
13
- 80.24426167704912
 
 
14
  ],
15
  "val_acc": [
16
  78.128,
17
- 79.002
 
 
18
  ],
19
  "lr": [
20
  0.001,
21
- 0.0009755527298894294
 
 
22
  ]
23
  },
24
  "loss_components": {
25
  "contrastive": [
26
  1.8756275988234499,
27
- 1.3881891413618581
 
 
28
  ],
29
  "correct": [
30
  0.6551205533000227,
31
- 0.550778582097052
 
 
32
  ],
33
  "incorrect": [
34
  0.46124604180121953,
35
- 0.4932183206747896
 
 
36
  ],
37
  "contrast": [
38
  1.979768048174465,
39
- 1.1816027985213282
 
 
40
  ],
41
  "coherence": [
42
  0.17198114866320263,
43
- 0.13845066018725613
 
 
44
  ],
45
  "separation": [
46
  0.019339063109397293,
47
- 0.03466135845487725
 
 
48
  ],
49
  "discretization": [
50
  0.10291576035837624,
51
- 0.058985046653858335
 
 
52
  ],
53
  "total": [
54
  1.9880030441779297,
55
- 1.4796098911533722
 
 
56
  ]
57
  },
58
  "scale_accuracies": {
59
  "576": [
60
  77.926,
61
- 78.328
 
 
62
  ],
63
  "640": [
64
  77.962,
65
- 78.396
 
 
66
  ],
67
  "704": [
68
  77.882,
69
- 78.294
 
 
70
  ],
71
  "768": [
72
  77.498,
73
- 78.442
 
 
74
  ],
75
  "832": [
76
  77.682,
77
- 78.292
 
 
78
  ],
79
  "896": [
80
  77.324,
81
- 77.498
 
 
82
  ],
83
  "960": [
84
  77.12,
85
- 76.88
 
 
86
  ]
87
  },
88
  "alpha_history": [
89
  0.4817016124725342,
90
- 0.414915531873703
 
 
91
  ]
92
  }
 
2
  "training_history": {
3
  "epochs": [
4
  1,
5
+ 2,
6
+ 3,
7
+ 4
8
  ],
9
  "train_loss": [
10
  1.9880030441779297,
11
+ 1.4796098911533722,
12
+ 1.3439062684298324,
13
+ 1.2507981145724703
14
  ],
15
  "train_acc": [
16
  74.1779955306373,
17
+ 80.24426167704912,
18
+ 81.11752800376532,
19
+ 81.97502745543711
20
  ],
21
  "val_acc": [
22
  78.128,
23
+ 79.002,
24
+ 79.918,
25
+ 80.514
26
  ],
27
  "lr": [
28
  0.001,
29
+ 0.0009755527298894294,
30
+ 0.0009046039886902864,
31
+ 0.0007940987335200905
32
  ]
33
  },
34
  "loss_components": {
35
  "contrastive": [
36
  1.8756275988234499,
37
+ 1.3881891413618581,
38
+ 1.2608741837949418,
39
+ 1.172699787365362
40
  ],
41
  "correct": [
42
  0.6551205533000227,
43
+ 0.550778582097052,
44
+ 0.4591032592490458,
45
+ 0.4241461445110294
46
  ],
47
  "incorrect": [
48
  0.46124604180121953,
49
+ 0.4932183206747896,
50
+ 0.5736374648900839,
51
+ 0.6029489130353014
52
  ],
53
  "contrast": [
54
  1.979768048174465,
55
+ 1.1816027985213282,
56
+ 1.0299043846777833,
57
+ 0.8941583722448958
58
  ],
59
  "coherence": [
60
  0.17198114866320263,
61
+ 0.13845066018725613,
62
+ 0.12057592163975256,
63
+ 0.10525317076170308
64
  ],
65
  "separation": [
66
  0.019339063109397293,
67
+ 0.03466135845487725,
68
+ 0.03344107698351621,
69
+ 0.026012230591380085
70
  ],
71
  "discretization": [
72
  0.10291576035837624,
73
+ 0.058985046653858335,
74
+ 0.06355896938145025,
75
+ 0.0883403262195727
76
  ],
77
  "total": [
78
  1.9880030441779297,
79
+ 1.4796098911533722,
80
+ 1.3439062684298324,
81
+ 1.2507981145724703
82
  ]
83
  },
84
  "scale_accuracies": {
85
  "576": [
86
  77.926,
87
+ 78.328,
88
+ 77.89,
89
+ 76.822
90
  ],
91
  "640": [
92
  77.962,
93
+ 78.396,
94
+ 78.004,
95
+ 76.962
96
  ],
97
  "704": [
98
  77.882,
99
+ 78.294,
100
+ 77.962,
101
+ 76.888
102
  ],
103
  "768": [
104
  77.498,
105
+ 78.442,
106
+ 78.888,
107
+ 79.468
108
  ],
109
  "832": [
110
  77.682,
111
+ 78.292,
112
+ 77.63,
113
+ 76.108
114
  ],
115
  "896": [
116
  77.324,
117
+ 77.498,
118
+ 74.308,
119
+ 70.584
120
  ],
121
  "960": [
122
  77.12,
123
+ 76.88,
124
+ 73.07,
125
+ 68.244
126
  ]
127
  },
128
  "alpha_history": [
129
  0.4817016124725342,
130
+ 0.414915531873703,
131
+ 0.36019453406333923,
132
+ 0.3008934259414673
133
  ]
134
  }