Epoch 50: 33.38%

README.md

@@ -5,7 +5,7 @@ tags:
 - cifar100
 - geometric-learning
 - fractal-encoding
-- trained
+- in-training
 - no-attention
 - no-cross-entropy
 datasets:
@@ -15,7 +15,7 @@ metrics:
 library_name: pytorch
 pipeline_tag: image-classification
 model-index:
-- name: geo-beatrix-resnet34-step18-feats1000
+- name: geo-beatrix-resnet34-step12-feats100
   results:
   - task:
       type: image-classification
@@ -30,13 +30,13 @@ model-index:
       verified: false
 ---
 
-# geo-beatrix-resnet34-step18-feats1000
+# geo-beatrix-resnet34-step12-feats100
 
 **Geometric Basin Classification for CIFAR-100**
 
-🎉 **Training Complete** 🎉
+🚧 **Training in Progress** 🚧
 
-Final Status: Epoch 200/200
+Current Status: Epoch 50/200
 
 ---
 
@@ -46,11 +46,11 @@ Final Status: Epoch 200/200
 |--------|-------|
 | **Best Test Accuracy** | **56.12%** |
 | **Best Epoch** | 160 |
-| **Current Train Accuracy** | 64.44% |
-| **Current Test Accuracy** | 53.62% |
-| **Current α (Cantor param)** | 0.4419 |
-| **Total Parameters** | 27,846,901 |
-| **Training Time** | 0:27:37 |
+| **Current Train Accuracy** | 56.74% |
+| **Current Test Accuracy** | 33.38% |
+| **Current α (Cantor param)** | 0.4995 |
+| **Total Parameters** | 21,846,001 |
+| **Training Time** | 0:02:59 |
 
 ### All Training Runs
 
@@ -59,6 +59,7 @@ Final Status: Epoch 200/200
 | `20251010_203717` | ✅ | 160 | **56.12%** | 67.82% | 0.4481 |
 | `20251010_211210` | 🔄 | 160 | **56.12%** | 16.21% | 0.3879 |
 | `20251010_213807` | ✅ | 160 | **56.12%** | 64.44% | 0.4419 |
+| `20251010_230300` | 🔄 | 160 | **56.12%** | 56.74% | 0.4995 |
 | `20251010_200842` | ✅ | 180 | **53.61%** | 67.53% | 0.4442 |
 | `20251010_185133` | ✅ | 200 | **52.97%** | 69.87% | 0.4452 |
 
@@ -66,8 +67,8 @@ Final Status: Epoch 200/200
 
 | Model | Accuracy | Status |
 |-------|----------|--------|
-| **geo-beatrix (this model)** | **56.12%** | ✅ Complete |
-| geo-beatrix      | 69.0% | Geometric Basin CONV architecture - 50m Params |
+| **geo-beatrix (this model)** | **56.12%** | 🔄 Training |
+| geo-beatrix (50M params) | 69.0% | Geometric Basin CONV architecture |
 
 🎯 **Current target**: Beat geo-beatrix (69.0%) - Currently -12.88%
 
@@ -75,12 +76,12 @@ Final Status: Epoch 200/200
 
 ## Architecture
 
-- **Base**: ResNet18 (torchvision)
+- **Base**: ResNet34 (torchvision)
 - **Pretrained**: From scratch
-- **Features**: 512-dim from ResNet18
+- **Features**: 512-dim from ResNet34
 - **Positional Encoding**: Devil's Staircase (Cantor function, 1883)
-- **PE Levels**: 18
-- **PE Features/Level**: 1000
+- **PE Levels**: 12
+- **PE Features/Level**: 100
 - **Classification**: Geometric Basin Compatibility (NO cross-entropy)
 - **Attention Mechanisms**: NONE
 - **Mixing**: Fractal (triadic multi-patch)
@@ -91,16 +92,16 @@ Final Status: Epoch 200/200
 
 ```json
 {
-  "model_name": "geo-beatrix-resnet34-step18-feats1000",
+  "model_name": "geo-beatrix-resnet34-step12-feats100",
   "model_type": "geometric_basin_classifier",
   "num_classes": 100,
-  "batch_size": 512,
+  "batch_size": 1024,
   "num_epochs": 200,
-  "base_learning_rate": 0.002,
-  "weight_decay": 0.05,
+  "base_learning_rate": 0.01,
+  "weight_decay": 0.0,
   "warmup_epochs": 10,
-  "pe_levels": 18,
-  "pe_features_per_level": 1000,
+  "pe_levels": 12,
+  "pe_features_per_level": 100,
   "dropout": 0.1,
   "pretrained_resnet": false,
   "frozen_resnet": false,
@@ -117,8 +118,8 @@ Final Status: Epoch 200/200
       0.3,
       0.7
     ],
-    "spatial_ratio": 0.25,
-    "curriculum_start": 0.0,
+    "spatial_ratio": 0.05,
+    "curriculum_start": 0.25,
     "curriculum_end": 0.5,
     "fractal_steps": [
       1,
@@ -130,11 +131,11 @@ Final Status: Epoch 200/200
       0.037037037037037035
     ]
   },
-  "architecture": "ResNet18 + Beatrix's Staircase PE",
+  "architecture": "ResNet34 + Devil's Staircase PE",
   "loss_function": "Geometric Basin Compatibility",
   "cross_entropy": false,
   "attention_mechanisms": false,
-  "timestamp": "20251010_213807"
+  "timestamp": "20251010_230300"
 }
 ```
 
@@ -148,7 +149,7 @@ Final Status: Epoch 200/200
 ├── best_model_info.json     (which epoch/run this came from)
 ├── runs_history.json        (all training runs and their results)
 ├── README.md
-├── weights/geo-beatrix-resnet34-step18-feats1000/20251010_213807/
+├── weights/geo-beatrix-resnet34-step12-feats100/20251010_230300/
 │   ├── model.pt                 (best from this training run)
 │   ├── model.safetensors        (best from this training run)
 │   ├── config.json
@@ -158,7 +159,7 @@ Final Status: Epoch 200/200
 │       ├── checkpoint_epoch_100.safetensors
 │       └── checkpoint_epoch_150.safetensors
 │       (snapshots every 10 epochs)
-└── runs/geo-beatrix-resnet34-step18-feats1000/20251010_213807/
+└── runs/geo-beatrix-resnet34-step12-feats100/20251010_230300/
     ├── events.out.tfevents.*    (TensorBoard logs)
     └── metrics.csv              (training metrics)
 ```
@@ -194,13 +195,13 @@ with open(info_path) as f:
 # Or download from specific training run
 model_path = hf_hub_download(
     repo_id="AbstractPhil/geo-beatrix-resnet",
-    filename="weights/geo-beatrix-resnet34-step18-feats1000/20251010_213807/model.safetensors"
+    filename="weights/geo-beatrix-resnet34-step12-feats100/20251010_230300/model.safetensors"
 )
 
 # Download specific epoch checkpoint
 epoch_checkpoint = hf_hub_download(
     repo_id="AbstractPhil/geo-beatrix-resnet",
-    filename="weights/geo-beatrix-resnet34-step18-feats1000/20251010_213807/checkpoints/checkpoint_epoch_100.safetensors"
+    filename="weights/geo-beatrix-resnet34-step12-feats100/20251010_230300/checkpoints/checkpoint_epoch_100.safetensors"
 )
 ```
 
@@ -210,23 +211,23 @@ epoch_checkpoint = hf_hub_download(
 
 ### Best Checkpoint
 - Epoch: 160
-- Train Acc: 65.96%
-- Test Acc: 54.14%
-- Alpha: 0.4427
-- Loss: 0.8208
+- Train Acc: 56.74%
+- Test Acc: 56.12%
+- Alpha: 0.4995
+- Loss: 0.0000
 
 ### Latest 5 Epochs
 
-- **Epoch 196**: Train 65.52%, Test 0.00%, α=0.4420, Loss=0.7704
-- **Epoch 197**: Train 62.64%, Test 0.00%, α=0.4420, Loss=0.7315
-- **Epoch 198**: Train 64.22%, Test 0.00%, α=0.4419, Loss=0.7351
-- **Epoch 199**: Train 60.50%, Test 0.00%, α=0.4419, Loss=0.7178
-- **Epoch 200**: Train 64.44%, Test 53.62%, α=0.4419, Loss=0.7510
+- **Epoch 46**: Train 60.77%, Test 33.07%, α=0.4970, Loss=0.9218
+- **Epoch 47**: Train 60.55%, Test 0.00%, α=0.5002, Loss=0.9846
+- **Epoch 48**: Train 57.74%, Test 33.91%, α=0.5023, Loss=0.9368
+- **Epoch 49**: Train 56.59%, Test 0.00%, α=0.4993, Loss=0.9326
+- **Epoch 50**: Train 56.74%, Test 33.38%, α=0.4995, Loss=0.8828
 
 ### Training Milestones
-- 🎯 **50% Accuracy** reached at epoch 85
-- 📊 **α ≥ 0.40** reached at epoch 12
-- 📊 **α ≥ 0.44** (near triadic equilibrium) at epoch 64
+- 📊 **α ≥ 0.40** reached at epoch 6
+- 📊 **α ≥ 0.44** (near triadic equilibrium) at epoch 8
+- ⚛️  **α = 0.50** (TRIADIC EQUILIBRIUM!) at epoch 21
 
 ---
 
@@ -236,7 +237,7 @@ epoch_checkpoint = hf_hub_download(
 ✅ **NO cross-entropy loss**  
 ✅ **Fractal positional encoding** (Cantor function from 1883)  
 ✅ **Geometric compatibility classification**  
-✅ **ResNet18 backbone** (proven CNN architecture)
+✅ **ResNet34 backbone** (proven CNN architecture)
 ✅ **Triadic fractal mixing** (base-3 aligned)
 
 ---