Epoch 50: 43.93%

README.md

@@ -15,7 +15,7 @@ metrics:
 library_name: pytorch
 pipeline_tag: image-classification
 model-index:
-- name: geo-beatrix-resnet18-imagenetpretrain-step18-feats1000
+- name: geo-beatrix-resnet34-step18-feats1000
   results:
   - task:
       type: image-classification
@@ -30,7 +30,7 @@ model-index:
       verified: false
 ---
 
-# geo-beatrix-resnet18-imagenetpretrain-step18-feats1000
+# geo-beatrix-resnet34-step18-feats1000
 
 **Geometric Basin Classification for CIFAR-100**
 
@@ -46,11 +46,11 @@ Current Status: Epoch 50/200
 |--------|-------|
 | **Best Test Accuracy** | **56.12%** |
 | **Best Epoch** | 160 |
-| **Current Train Accuracy** | 16.21% |
-| **Current Test Accuracy** | 17.03% |
-| **Current α (Cantor param)** | 0.3879 |
-| **Total Parameters** | 17,738,741 |
-| **Training Time** | 0:06:31 |
+| **Current Train Accuracy** | 42.89% |
+| **Current Test Accuracy** | 43.93% |
+| **Current α (Cantor param)** | 0.4310 |
+| **Total Parameters** | 27,846,901 |
+| **Training Time** | 0:06:50 |
 
 ### All Training Runs
 
@@ -58,6 +58,7 @@ Current Status: Epoch 50/200
 |-----------|--------|------------|----------|-----------|---|
 | `20251010_203717` | ✅ | 160 | **56.12%** | 67.82% | 0.4481 |
 | `20251010_211210` | 🔄 | 160 | **56.12%** | 16.21% | 0.3879 |
+| `20251010_213807` | 🔄 | 160 | **56.12%** | 42.89% | 0.4310 |
 | `20251010_200842` | ✅ | 180 | **53.61%** | 67.53% | 0.4442 |
 | `20251010_185133` | ✅ | 200 | **52.97%** | 69.87% | 0.4452 |
 
@@ -75,7 +76,7 @@ Current Status: Epoch 50/200
 ## Architecture
 
 - **Base**: ResNet18 (torchvision)
-- **Pretrained**: ImageNet
+- **Pretrained**: From scratch
 - **Features**: 512-dim from ResNet18
 - **Positional Encoding**: Devil's Staircase (Cantor function, 1883)
 - **PE Levels**: 18
@@ -90,7 +91,7 @@ Current Status: Epoch 50/200
 
 ```json
 {
-  "model_name": "geo-beatrix-resnet18-imagenetpretrain-step18-feats1000",
+  "model_name": "geo-beatrix-resnet34-step18-feats1000",
   "model_type": "geometric_basin_classifier",
   "num_classes": 100,
   "batch_size": 512,
@@ -101,8 +102,8 @@ Current Status: Epoch 50/200
   "pe_levels": 18,
   "pe_features_per_level": 1000,
   "dropout": 0.1,
-  "pretrained_resnet": true,
-  "frozen_resnet": true,
+  "pretrained_resnet": false,
+  "frozen_resnet": false,
   "a100_optimizations": {
     "mixed_precision": true,
     "torch_compile": false,
@@ -133,7 +134,7 @@ Current Status: Epoch 50/200
   "loss_function": "Geometric Basin Compatibility",
   "cross_entropy": false,
   "attention_mechanisms": false,
-  "timestamp": "20251010_211210"
+  "timestamp": "20251010_213807"
 }
 ```
 
@@ -147,7 +148,7 @@ Current Status: Epoch 50/200
 ├── best_model_info.json     (which epoch/run this came from)
 ├── runs_history.json        (all training runs and their results)
 ├── README.md
-├── weights/geo-beatrix-resnet18-imagenetpretrain-step18-feats1000/20251010_211210/
+├── weights/geo-beatrix-resnet34-step18-feats1000/20251010_213807/
 │   ├── model.pt                 (best from this training run)
 │   ├── model.safetensors        (best from this training run)
 │   ├── config.json
@@ -157,7 +158,7 @@ Current Status: Epoch 50/200
 │       ├── checkpoint_epoch_100.safetensors
 │       └── checkpoint_epoch_150.safetensors
 │       (snapshots every 10 epochs)
-└── runs/geo-beatrix-resnet18-imagenetpretrain-step18-feats1000/20251010_211210/
+└── runs/geo-beatrix-resnet34-step18-feats1000/20251010_213807/
     ├── events.out.tfevents.*    (TensorBoard logs)
     └── metrics.csv              (training metrics)
 ```
@@ -193,13 +194,13 @@ 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-resnet18-imagenetpretrain-step18-feats1000/20251010_211210/model.safetensors"
+    filename="weights/geo-beatrix-resnet34-step18-feats1000/20251010_213807/model.safetensors"
 )
 
 # Download specific epoch checkpoint
 epoch_checkpoint = hf_hub_download(
     repo_id="AbstractPhil/geo-beatrix-resnet",
-    filename="weights/geo-beatrix-resnet18-imagenetpretrain-step18-feats1000/20251010_211210/checkpoints/checkpoint_epoch_100.safetensors"
+    filename="weights/geo-beatrix-resnet34-step18-feats1000/20251010_213807/checkpoints/checkpoint_epoch_100.safetensors"
 )
 ```
 
@@ -209,22 +210,21 @@ epoch_checkpoint = hf_hub_download(
 
 ### Best Checkpoint
 - Epoch: 160
-- Train Acc: 16.21%
+- Train Acc: 42.89%
 - Test Acc: 56.12%
-- Alpha: 0.3879
+- Alpha: 0.4310
 - Loss: 0.0000
 
 ### Latest 5 Epochs
 
-- **Epoch 46**: Train 16.06%, Test 0.00%, α=0.3940, Loss=2.2744
-- **Epoch 47**: Train 16.18%, Test 0.00%, α=0.3953, Loss=2.3605
-- **Epoch 48**: Train 16.25%, Test 0.00%, α=0.3902, Loss=2.2532
-- **Epoch 49**: Train 16.27%, Test 0.00%, α=0.3907, Loss=2.2053
-- **Epoch 50**: Train 16.21%, Test 17.03%, α=0.3879, Loss=2.2668
+- **Epoch 46**: Train 41.81%, Test 0.00%, α=0.4305, Loss=1.6292
+- **Epoch 47**: Train 42.20%, Test 0.00%, α=0.4324, Loss=1.5935
+- **Epoch 48**: Train 43.09%, Test 0.00%, α=0.4302, Loss=1.6075
+- **Epoch 49**: Train 42.97%, Test 0.00%, α=0.4346, Loss=1.5375
+- **Epoch 50**: Train 42.89%, Test 43.93%, α=0.4310, Loss=1.4858
 
 ### Training Milestones
-
-- 🔄 Frozen complete failure. Resnet18 concluded.
+- 📊 **α ≥ 0.40** reached at epoch 12
 
 ---