Update Gyuyig vs Tsugdri binary classifier: center_crop 224 metrics, confusion matrix, training history
Browse files- README.md +29 -29
- config.yaml +6 -5
- confusion_matrix.json +19 -19
- confusion_matrix.png +0 -0
- final_model.pt +1 -1
- model_card.json +21 -21
- results.json +174 -240
- training_history.png +0 -0
README.md
CHANGED
|
@@ -25,11 +25,11 @@ Fine-tuned [DINOv3 ViT-S](https://huggingface.co/facebook/dinov3-vits16-pretrain
|
|
| 25 |
|
| 26 |
**Gyuyig**, **Tsugdri**
|
| 27 |
|
| 28 |
-
**Experiment:** `
|
| 29 |
**Pooling:** ViT **CLS token** (`last_hidden_state[:, 0, :]`)
|
| 30 |
**Weights:** `final_model.pt` (best validation macro-F1 across stages A/B/C)
|
| 31 |
|
| 32 |
-
**Warm-start:** [BDRC/4-class-balanced-script-classifier](https://huggingface.co/BDRC/4-class-balanced-script-classifier) (`final_model.pt` — prior test acc
|
| 33 |
|
| 34 |
## Data
|
| 35 |
|
|
@@ -43,56 +43,56 @@ Test split: balanced benchmark (60 images per parent class, held out of training
|
|
| 43 |
|
| 44 |
| Split | Mode | Size |
|
| 45 |
|-------|------|-----:|
|
| 46 |
-
| train | `
|
| 47 |
-
| val | `
|
| 48 |
-
| test | `
|
| 49 |
|
| 50 |
## Validation metrics (n=60)
|
| 51 |
|
| 52 |
| Metric | Value |
|
| 53 |
|--------|------:|
|
| 54 |
-
| Accuracy |
|
| 55 |
-
| Macro F1 | 0.
|
| 56 |
-
| Weighted F1 | 0.
|
| 57 |
-
| AUC-ROC | 0.
|
| 58 |
-
| Loss | 0.
|
| 59 |
|
| 60 |
-
**Best checkpoint:** `best_stage_c_last_blocks.pt` epoch
|
| 61 |
|
| 62 |
### Per-class (validation)
|
| 63 |
|
| 64 |
```
|
| 65 |
precision recall f1-score support
|
| 66 |
|
| 67 |
-
Gyuyig 0.
|
| 68 |
-
Tsugdri 0.
|
| 69 |
|
| 70 |
-
accuracy 0.
|
| 71 |
-
macro avg 0.
|
| 72 |
-
weighted avg 0.
|
| 73 |
```
|
| 74 |
|
| 75 |
## Test / benchmark metrics (n=120)
|
| 76 |
|
| 77 |
| Metric | Value |
|
| 78 |
|--------|------:|
|
| 79 |
-
| Accuracy |
|
| 80 |
-
| Macro F1 | 0.
|
| 81 |
-
| Weighted F1 | 0.
|
| 82 |
-
| AUC-ROC | 0.
|
| 83 |
-
| Loss | 0.
|
| 84 |
|
| 85 |
### Per-class (test)
|
| 86 |
|
| 87 |
```
|
| 88 |
precision recall f1-score support
|
| 89 |
|
| 90 |
-
Gyuyig 0.
|
| 91 |
-
Tsugdri 0.
|
| 92 |
|
| 93 |
-
accuracy 0.
|
| 94 |
-
macro avg 0.
|
| 95 |
-
weighted avg 0.
|
| 96 |
```
|
| 97 |
|
| 98 |
## Training
|
|
@@ -116,8 +116,8 @@ weighted avg 0.81 0.81 0.81 120
|
|
| 116 |
|
| 117 |
| True \ Pred | Gyuyig | Tsugdri |
|
| 118 |
|---|---:|---:|
|
| 119 |
-
| **Gyuyig** |
|
| 120 |
-
| **Tsugdri** |
|
| 121 |
|
| 122 |
## Files
|
| 123 |
|
|
@@ -137,7 +137,7 @@ weighted avg 0.81 0.81 0.81 120
|
|
| 137 |
|
| 138 |
```bash
|
| 139 |
pip install -r requirements-inference.txt
|
| 140 |
-
python inference.py --checkpoint final_model.pt --image path/to/page.jpg --preprocess resize_letterbox --preprocess-size
|
| 141 |
```
|
| 142 |
|
| 143 |
## Reproduce training
|
|
|
|
| 25 |
|
| 26 |
**Gyuyig**, **Tsugdri**
|
| 27 |
|
| 28 |
+
**Experiment:** `dinov3_gyuyig_tsugdri_binary` (`gyuyig_tsugdri_binary_classification`)
|
| 29 |
**Pooling:** ViT **CLS token** (`last_hidden_state[:, 0, :]`)
|
| 30 |
**Weights:** `final_model.pt` (best validation macro-F1 across stages A/B/C)
|
| 31 |
|
| 32 |
+
**Warm-start:** [BDRC/4-class-balanced-script-classifier](https://huggingface.co/BDRC/4-class-balanced-script-classifier) (`final_model.pt` — prior test acc 92.1%, macro-F1 0.921)
|
| 33 |
|
| 34 |
## Data
|
| 35 |
|
|
|
|
| 43 |
|
| 44 |
| Split | Mode | Size |
|
| 45 |
|-------|------|-----:|
|
| 46 |
+
| train | `center_crop` | 224 |
|
| 47 |
+
| val | `center_crop` | 224 |
|
| 48 |
+
| test | `center_crop` | 224 |
|
| 49 |
|
| 50 |
## Validation metrics (n=60)
|
| 51 |
|
| 52 |
| Metric | Value |
|
| 53 |
|--------|------:|
|
| 54 |
+
| Accuracy | 91.7% |
|
| 55 |
+
| Macro F1 | 0.916 |
|
| 56 |
+
| Weighted F1 | 0.916 |
|
| 57 |
+
| AUC-ROC | 0.931 |
|
| 58 |
+
| Loss | 0.3915 |
|
| 59 |
|
| 60 |
+
**Best checkpoint:** `best_stage_c_last_blocks.pt` epoch 1 val macro-F1 0.916
|
| 61 |
|
| 62 |
### Per-class (validation)
|
| 63 |
|
| 64 |
```
|
| 65 |
precision recall f1-score support
|
| 66 |
|
| 67 |
+
Gyuyig 0.88 0.97 0.92 30
|
| 68 |
+
Tsugdri 0.96 0.87 0.91 30
|
| 69 |
|
| 70 |
+
accuracy 0.92 60
|
| 71 |
+
macro avg 0.92 0.92 0.92 60
|
| 72 |
+
weighted avg 0.92 0.92 0.92 60
|
| 73 |
```
|
| 74 |
|
| 75 |
## Test / benchmark metrics (n=120)
|
| 76 |
|
| 77 |
| Metric | Value |
|
| 78 |
|--------|------:|
|
| 79 |
+
| Accuracy | 85.0% |
|
| 80 |
+
| Macro F1 | 0.848 |
|
| 81 |
+
| Weighted F1 | 0.848 |
|
| 82 |
+
| AUC-ROC | 0.930 |
|
| 83 |
+
| Loss | 0.4047 |
|
| 84 |
|
| 85 |
### Per-class (test)
|
| 86 |
|
| 87 |
```
|
| 88 |
precision recall f1-score support
|
| 89 |
|
| 90 |
+
Gyuyig 0.78 0.97 0.87 60
|
| 91 |
+
Tsugdri 0.96 0.73 0.83 60
|
| 92 |
|
| 93 |
+
accuracy 0.85 120
|
| 94 |
+
macro avg 0.87 0.85 0.85 120
|
| 95 |
+
weighted avg 0.87 0.85 0.85 120
|
| 96 |
```
|
| 97 |
|
| 98 |
## Training
|
|
|
|
| 116 |
|
| 117 |
| True \ Pred | Gyuyig | Tsugdri |
|
| 118 |
|---|---:|---:|
|
| 119 |
+
| **Gyuyig** | 58 | 2 |
|
| 120 |
+
| **Tsugdri** | 16 | 44 |
|
| 121 |
|
| 122 |
## Files
|
| 123 |
|
|
|
|
| 137 |
|
| 138 |
```bash
|
| 139 |
pip install -r requirements-inference.txt
|
| 140 |
+
python inference.py --checkpoint final_model.pt --image path/to/page.jpg --preprocess resize_letterbox --preprocess-size 224
|
| 141 |
```
|
| 142 |
|
| 143 |
## Reproduce training
|
config.yaml
CHANGED
|
@@ -1,9 +1,10 @@
|
|
| 1 |
-
experiment:
|
| 2 |
task: gyuyig_tsugdri_binary_classification
|
| 3 |
|
| 4 |
balanced_dataset_repo: BDRC/gyuyig-tsugdri-binary-balanced-script-classification-dataset
|
| 5 |
val_ratio: 0.15
|
| 6 |
|
|
|
|
| 7 |
warmstart_repo: BDRC/4-class-balanced-script-classifier
|
| 8 |
warmstart_checkpoint_file: final_model.pt
|
| 9 |
|
|
@@ -17,10 +18,10 @@ no_weighted_sampler: true
|
|
| 17 |
skip_stage_c: false
|
| 18 |
gradient_checkpointing: true
|
| 19 |
|
| 20 |
-
train_preprocess:
|
| 21 |
-
val_preprocess:
|
| 22 |
-
test_preprocess:
|
| 23 |
-
preprocess_size:
|
| 24 |
|
| 25 |
pooling: cls_token
|
| 26 |
|
|
|
|
| 1 |
+
experiment: dinov3_gyuyig_tsugdri_binary
|
| 2 |
task: gyuyig_tsugdri_binary_classification
|
| 3 |
|
| 4 |
balanced_dataset_repo: BDRC/gyuyig-tsugdri-binary-balanced-script-classification-dataset
|
| 5 |
val_ratio: 0.15
|
| 6 |
|
| 7 |
+
|
| 8 |
warmstart_repo: BDRC/4-class-balanced-script-classifier
|
| 9 |
warmstart_checkpoint_file: final_model.pt
|
| 10 |
|
|
|
|
| 18 |
skip_stage_c: false
|
| 19 |
gradient_checkpointing: true
|
| 20 |
|
| 21 |
+
train_preprocess: center_crop
|
| 22 |
+
val_preprocess: center_crop
|
| 23 |
+
test_preprocess: center_crop
|
| 24 |
+
preprocess_size: 224
|
| 25 |
|
| 26 |
pooling: cls_token
|
| 27 |
|
confusion_matrix.json
CHANGED
|
@@ -6,36 +6,36 @@
|
|
| 6 |
],
|
| 7 |
"matrix": [
|
| 8 |
[
|
| 9 |
-
|
| 10 |
-
|
| 11 |
],
|
| 12 |
[
|
| 13 |
-
|
| 14 |
-
|
| 15 |
]
|
| 16 |
],
|
| 17 |
"test_metrics": {
|
| 18 |
-
"loss": 0.
|
| 19 |
-
"accuracy": 0.
|
| 20 |
-
"macro_f1": 0.
|
| 21 |
-
"weighted_f1": 0.
|
| 22 |
-
"auc_roc": 0.
|
| 23 |
},
|
| 24 |
"val_metrics": {
|
| 25 |
-
"loss": 0.
|
| 26 |
-
"accuracy": 0.
|
| 27 |
-
"macro_f1": 0.
|
| 28 |
-
"weighted_f1": 0.
|
| 29 |
-
"auc_roc": 0.
|
| 30 |
},
|
| 31 |
"preprocess": {
|
| 32 |
-
"train": "
|
| 33 |
-
"val": "
|
| 34 |
-
"test": "
|
| 35 |
-
"size":
|
| 36 |
},
|
| 37 |
"train_dataset": "BDRC/gyuyig-tsugdri-binary-balanced-script-classification-dataset",
|
| 38 |
"benchmark_per_parent": 60,
|
| 39 |
-
"experiment": "
|
| 40 |
"repo_id": "BDRC/gyuyig-tsugdri-binary-script-classifier"
|
| 41 |
}
|
|
|
|
| 6 |
],
|
| 7 |
"matrix": [
|
| 8 |
[
|
| 9 |
+
58,
|
| 10 |
+
2
|
| 11 |
],
|
| 12 |
[
|
| 13 |
+
16,
|
| 14 |
+
44
|
| 15 |
]
|
| 16 |
],
|
| 17 |
"test_metrics": {
|
| 18 |
+
"loss": 0.40474860469500223,
|
| 19 |
+
"accuracy": 0.85,
|
| 20 |
+
"macro_f1": 0.847930160518164,
|
| 21 |
+
"weighted_f1": 0.847930160518164,
|
| 22 |
+
"auc_roc": 0.9297222222222223
|
| 23 |
},
|
| 24 |
"val_metrics": {
|
| 25 |
+
"loss": 0.3914561231931051,
|
| 26 |
+
"accuracy": 0.9166666666666666,
|
| 27 |
+
"macro_f1": 0.9164578111946533,
|
| 28 |
+
"weighted_f1": 0.9164578111946533,
|
| 29 |
+
"auc_roc": 0.9311111111111111
|
| 30 |
},
|
| 31 |
"preprocess": {
|
| 32 |
+
"train": "center_crop",
|
| 33 |
+
"val": "center_crop",
|
| 34 |
+
"test": "center_crop",
|
| 35 |
+
"size": 224
|
| 36 |
},
|
| 37 |
"train_dataset": "BDRC/gyuyig-tsugdri-binary-balanced-script-classification-dataset",
|
| 38 |
"benchmark_per_parent": 60,
|
| 39 |
+
"experiment": "dinov3_gyuyig_tsugdri_binary",
|
| 40 |
"repo_id": "BDRC/gyuyig-tsugdri-binary-script-classifier"
|
| 41 |
}
|
confusion_matrix.png
CHANGED
|
|
final_model.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 86670182
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9cce9a6920cff9cd5fc7add0a40b81969f36383fc1919fcb4c0adb1b0f2047b1
|
| 3 |
size 86670182
|
model_card.json
CHANGED
|
@@ -3,27 +3,27 @@
|
|
| 3 |
"train_dataset_id": "BDRC/gyuyig-tsugdri-binary-balanced-script-classification-dataset",
|
| 4 |
"warmstart_repo": "BDRC/4-class-balanced-script-classifier",
|
| 5 |
"task": "gyuyig_tsugdri_binary_classification",
|
| 6 |
-
"experiment": "
|
| 7 |
"classes": [
|
| 8 |
"Gyuyig",
|
| 9 |
"Tsugdri"
|
| 10 |
],
|
| 11 |
"pooling": "cls_token",
|
| 12 |
"preprocess": {
|
| 13 |
-
"train": "
|
| 14 |
-
"val": "
|
| 15 |
-
"test": "
|
| 16 |
-
"size":
|
| 17 |
},
|
| 18 |
"warmstart": {
|
| 19 |
"warmstart_repo": "BDRC/4-class-balanced-script-classifier",
|
| 20 |
"warmstart_checkpoint": null,
|
| 21 |
"warmstart_checkpoint_file": "final_model.pt",
|
| 22 |
"checkpoint_test_metrics": {
|
| 23 |
-
"loss": 0.
|
| 24 |
-
"accuracy": 0.
|
| 25 |
-
"macro_f1": 0.
|
| 26 |
-
"weighted_f1": 0.
|
| 27 |
},
|
| 28 |
"warmstart_pooling": "cls_token"
|
| 29 |
},
|
|
@@ -66,21 +66,21 @@
|
|
| 66 |
},
|
| 67 |
"best_checkpoint": {
|
| 68 |
"path": "best_stage_c_last_blocks.pt",
|
| 69 |
-
"epoch":
|
| 70 |
-
"val_macro_f1": 0.
|
| 71 |
},
|
| 72 |
"val_metrics": {
|
| 73 |
-
"loss": 0.
|
| 74 |
-
"accuracy": 0.
|
| 75 |
-
"macro_f1": 0.
|
| 76 |
-
"weighted_f1": 0.
|
| 77 |
-
"auc_roc": 0.
|
| 78 |
},
|
| 79 |
"test_metrics": {
|
| 80 |
-
"loss": 0.
|
| 81 |
-
"accuracy": 0.
|
| 82 |
-
"macro_f1": 0.
|
| 83 |
-
"weighted_f1": 0.
|
| 84 |
-
"auc_roc": 0.
|
| 85 |
}
|
| 86 |
}
|
|
|
|
| 3 |
"train_dataset_id": "BDRC/gyuyig-tsugdri-binary-balanced-script-classification-dataset",
|
| 4 |
"warmstart_repo": "BDRC/4-class-balanced-script-classifier",
|
| 5 |
"task": "gyuyig_tsugdri_binary_classification",
|
| 6 |
+
"experiment": "dinov3_gyuyig_tsugdri_binary",
|
| 7 |
"classes": [
|
| 8 |
"Gyuyig",
|
| 9 |
"Tsugdri"
|
| 10 |
],
|
| 11 |
"pooling": "cls_token",
|
| 12 |
"preprocess": {
|
| 13 |
+
"train": "center_crop",
|
| 14 |
+
"val": "center_crop",
|
| 15 |
+
"test": "center_crop",
|
| 16 |
+
"size": 224
|
| 17 |
},
|
| 18 |
"warmstart": {
|
| 19 |
"warmstart_repo": "BDRC/4-class-balanced-script-classifier",
|
| 20 |
"warmstart_checkpoint": null,
|
| 21 |
"warmstart_checkpoint_file": "final_model.pt",
|
| 22 |
"checkpoint_test_metrics": {
|
| 23 |
+
"loss": 0.42663901050885517,
|
| 24 |
+
"accuracy": 0.9208333333333333,
|
| 25 |
+
"macro_f1": 0.9208489161207983,
|
| 26 |
+
"weighted_f1": 0.9208489161207983
|
| 27 |
},
|
| 28 |
"warmstart_pooling": "cls_token"
|
| 29 |
},
|
|
|
|
| 66 |
},
|
| 67 |
"best_checkpoint": {
|
| 68 |
"path": "best_stage_c_last_blocks.pt",
|
| 69 |
+
"epoch": 1,
|
| 70 |
+
"val_macro_f1": 0.9164578111946533
|
| 71 |
},
|
| 72 |
"val_metrics": {
|
| 73 |
+
"loss": 0.3914561231931051,
|
| 74 |
+
"accuracy": 0.9166666666666666,
|
| 75 |
+
"macro_f1": 0.9164578111946533,
|
| 76 |
+
"weighted_f1": 0.9164578111946533,
|
| 77 |
+
"auc_roc": 0.9311111111111111
|
| 78 |
},
|
| 79 |
"test_metrics": {
|
| 80 |
+
"loss": 0.40474860469500223,
|
| 81 |
+
"accuracy": 0.85,
|
| 82 |
+
"macro_f1": 0.847930160518164,
|
| 83 |
+
"weighted_f1": 0.847930160518164,
|
| 84 |
+
"auc_roc": 0.9297222222222223
|
| 85 |
}
|
| 86 |
}
|
results.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"experiment": "
|
| 3 |
"run_subdir": null,
|
| 4 |
"task": "gyuyig_tsugdri_binary_classification",
|
| 5 |
"balanced_parquet_dir": null,
|
|
@@ -8,10 +8,10 @@
|
|
| 8 |
"benchmark_dataset_repo": null,
|
| 9 |
"benchmark_per_parent": 60,
|
| 10 |
"preprocess": {
|
| 11 |
-
"train": "
|
| 12 |
-
"val": "
|
| 13 |
-
"test": "
|
| 14 |
-
"size":
|
| 15 |
},
|
| 16 |
"pooling": "cls_token",
|
| 17 |
"training_config": {
|
|
@@ -60,10 +60,10 @@
|
|
| 60 |
"warmstart_checkpoint": null,
|
| 61 |
"warmstart_checkpoint_file": "final_model.pt",
|
| 62 |
"checkpoint_test_metrics": {
|
| 63 |
-
"loss": 0.
|
| 64 |
-
"accuracy": 0.
|
| 65 |
-
"macro_f1": 0.
|
| 66 |
-
"weighted_f1": 0.
|
| 67 |
},
|
| 68 |
"warmstart_pooling": "cls_token"
|
| 69 |
},
|
|
@@ -72,376 +72,310 @@
|
|
| 72 |
},
|
| 73 |
"best_checkpoint": {
|
| 74 |
"path": "best_stage_c_last_blocks.pt",
|
| 75 |
-
"epoch":
|
| 76 |
-
"val_macro_f1": 0.
|
| 77 |
},
|
| 78 |
"val_metrics": {
|
| 79 |
-
"loss": 0.
|
| 80 |
-
"accuracy": 0.
|
| 81 |
-
"macro_f1": 0.
|
| 82 |
-
"weighted_f1": 0.
|
| 83 |
-
"auc_roc": 0.
|
| 84 |
},
|
| 85 |
"val_confusion_matrix": [
|
| 86 |
[
|
| 87 |
-
|
| 88 |
-
|
| 89 |
],
|
| 90 |
[
|
| 91 |
-
|
| 92 |
-
|
| 93 |
]
|
| 94 |
],
|
| 95 |
-
"val_report": " precision recall f1-score support\n\n Gyuyig 0.
|
| 96 |
"test_metrics": {
|
| 97 |
-
"loss": 0.
|
| 98 |
-
"accuracy": 0.
|
| 99 |
-
"macro_f1": 0.
|
| 100 |
-
"weighted_f1": 0.
|
| 101 |
-
"auc_roc": 0.
|
| 102 |
},
|
| 103 |
"test_confusion_matrix": [
|
| 104 |
[
|
| 105 |
-
|
| 106 |
-
|
| 107 |
],
|
| 108 |
[
|
| 109 |
-
|
| 110 |
-
|
| 111 |
]
|
| 112 |
],
|
| 113 |
-
"test_report": " precision recall f1-score support\n\n Gyuyig 0.
|
| 114 |
"history": {
|
| 115 |
"stage_a": [
|
| 116 |
{
|
| 117 |
"epoch": 1,
|
| 118 |
-
"train_loss": 0.
|
| 119 |
-
"train_acc": 0.
|
| 120 |
"lr_head": 5e-06,
|
| 121 |
-
"val_loss": 0.
|
| 122 |
-
"val_accuracy": 0.
|
| 123 |
-
"val_macro_f1": 0.
|
| 124 |
-
"val_weighted_f1": 0.
|
| 125 |
},
|
| 126 |
{
|
| 127 |
"epoch": 2,
|
| 128 |
-
"train_loss": 0.
|
| 129 |
-
"train_acc": 0.
|
| 130 |
"lr_head": 0.0005,
|
| 131 |
-
"val_loss": 0.
|
| 132 |
-
"val_accuracy": 0.
|
| 133 |
-
"val_macro_f1": 0.
|
| 134 |
-
"val_weighted_f1": 0.
|
| 135 |
},
|
| 136 |
{
|
| 137 |
"epoch": 3,
|
| 138 |
-
"train_loss": 0.
|
| 139 |
-
"train_acc": 0.
|
| 140 |
"lr_head": 0.0004668412874366486,
|
| 141 |
-
"val_loss": 0.
|
| 142 |
-
"val_accuracy": 0.
|
| 143 |
-
"val_macro_f1": 0.
|
| 144 |
-
"val_weighted_f1": 0.
|
| 145 |
},
|
| 146 |
{
|
| 147 |
"epoch": 4,
|
| 148 |
-
"train_loss": 0.
|
| 149 |
-
"train_acc": 0.
|
| 150 |
"lr_head": 0.00037624999999999996,
|
| 151 |
-
"val_loss": 0.
|
| 152 |
-
"val_accuracy": 0.
|
| 153 |
-
"val_macro_f1": 0.
|
| 154 |
-
"val_weighted_f1": 0.
|
| 155 |
},
|
| 156 |
{
|
| 157 |
"epoch": 5,
|
| 158 |
-
"train_loss": 0.
|
| 159 |
-
"train_acc": 0.
|
| 160 |
"lr_head": 0.0002525,
|
| 161 |
-
"val_loss": 0.
|
| 162 |
-
"val_accuracy": 0.
|
| 163 |
-
"val_macro_f1": 0.
|
| 164 |
-
"val_weighted_f1": 0.
|
| 165 |
},
|
| 166 |
{
|
| 167 |
"epoch": 6,
|
| 168 |
-
"train_loss": 0.
|
| 169 |
-
"train_acc": 0.
|
| 170 |
"lr_head": 0.00012875000000000007,
|
| 171 |
-
"val_loss": 0.
|
| 172 |
-
"val_accuracy": 0.
|
| 173 |
-
"val_macro_f1": 0.
|
| 174 |
-
"val_weighted_f1": 0.
|
| 175 |
},
|
| 176 |
{
|
| 177 |
"epoch": 7,
|
| 178 |
-
"train_loss": 0.
|
| 179 |
-
"train_acc": 0.
|
| 180 |
"lr_head": 3.815871256335142e-05,
|
| 181 |
-
"val_loss": 0.
|
| 182 |
-
"val_accuracy": 0.
|
| 183 |
-
"val_macro_f1": 0.
|
| 184 |
-
"val_weighted_f1": 0.
|
| 185 |
}
|
| 186 |
],
|
| 187 |
"stage_b": [
|
| 188 |
{
|
| 189 |
"epoch": 1,
|
| 190 |
-
"train_loss": 0.
|
| 191 |
-
"train_acc": 0.
|
| 192 |
"lr_head": 1.0000000000000002e-06,
|
| 193 |
"lr_backbone": 1.0000000000000001e-07,
|
| 194 |
-
"val_loss": 0.
|
| 195 |
-
"val_accuracy": 0.
|
| 196 |
-
"val_macro_f1": 0.
|
| 197 |
-
"val_weighted_f1": 0.
|
| 198 |
},
|
| 199 |
{
|
| 200 |
"epoch": 2,
|
| 201 |
-
"train_loss": 0.
|
| 202 |
-
"train_acc": 0.
|
| 203 |
"lr_head": 0.0001,
|
| 204 |
"lr_backbone": 1e-05,
|
| 205 |
-
"val_loss": 0.
|
| 206 |
-
"val_accuracy": 0.
|
| 207 |
-
"val_macro_f1": 0.
|
| 208 |
-
"val_weighted_f1": 0.
|
| 209 |
},
|
| 210 |
{
|
| 211 |
"epoch": 3,
|
| 212 |
-
"train_loss": 0.
|
| 213 |
-
"train_acc": 0.
|
| 214 |
"lr_head": 9.701478472890248e-05,
|
| 215 |
"lr_backbone": 9.701478472890248e-06,
|
| 216 |
-
"val_loss": 0.
|
| 217 |
-
"val_accuracy": 0.
|
| 218 |
-
"val_macro_f1": 0.
|
| 219 |
-
"val_weighted_f1": 0.
|
| 220 |
},
|
| 221 |
{
|
| 222 |
"epoch": 4,
|
| 223 |
-
"train_loss": 0.
|
| 224 |
-
"train_acc": 0.
|
| 225 |
"lr_head": 8.84191999343894e-05,
|
| 226 |
"lr_backbone": 8.841919993438941e-06,
|
| 227 |
-
"val_loss": 0.
|
| 228 |
-
"val_accuracy": 0.
|
| 229 |
-
"val_macro_f1": 0.
|
| 230 |
-
"val_weighted_f1": 0.
|
| 231 |
},
|
| 232 |
{
|
| 233 |
"epoch": 5,
|
| 234 |
-
"train_loss": 0.
|
| 235 |
-
"train_acc": 0.
|
| 236 |
"lr_head": 7.525e-05,
|
| 237 |
"lr_backbone": 7.525e-06,
|
| 238 |
-
"val_loss": 0.
|
| 239 |
-
"val_accuracy": 0.
|
| 240 |
-
"val_macro_f1": 0.
|
| 241 |
-
"val_weighted_f1": 0.
|
| 242 |
},
|
| 243 |
{
|
| 244 |
"epoch": 6,
|
| 245 |
-
"train_loss": 0.
|
| 246 |
-
"train_acc": 0.
|
| 247 |
"lr_head": 5.909558479451306e-05,
|
| 248 |
"lr_backbone": 5.909558479451306e-06,
|
| 249 |
-
"val_loss": 0.
|
| 250 |
-
"val_accuracy": 0.
|
| 251 |
-
"val_macro_f1": 0.
|
| 252 |
-
"val_weighted_f1": 0.
|
| 253 |
},
|
| 254 |
{
|
| 255 |
"epoch": 7,
|
| 256 |
-
"train_loss": 0.
|
| 257 |
-
"train_acc": 0.
|
| 258 |
"lr_head": 4.190441520548695e-05,
|
| 259 |
"lr_backbone": 4.190441520548696e-06,
|
| 260 |
-
"val_loss": 0.
|
| 261 |
-
"val_accuracy": 0.
|
| 262 |
-
"val_macro_f1": 0.
|
| 263 |
-
"val_weighted_f1": 0.
|
| 264 |
},
|
| 265 |
{
|
| 266 |
"epoch": 8,
|
| 267 |
-
"train_loss": 0.
|
| 268 |
-
"train_acc": 0.
|
| 269 |
"lr_head": 2.5750000000000013e-05,
|
| 270 |
"lr_backbone": 2.575000000000001e-06,
|
| 271 |
-
"val_loss": 0.
|
| 272 |
-
"val_accuracy": 0.
|
| 273 |
-
"val_macro_f1": 0.
|
| 274 |
-
"val_weighted_f1": 0.
|
| 275 |
},
|
| 276 |
{
|
| 277 |
"epoch": 9,
|
| 278 |
-
"train_loss": 0.
|
| 279 |
-
"train_acc": 0.
|
| 280 |
"lr_head": 1.2580800065610596e-05,
|
| 281 |
"lr_backbone": 1.2580800065610596e-06,
|
| 282 |
-
"val_loss": 0.
|
| 283 |
-
"val_accuracy": 0.
|
| 284 |
-
"val_macro_f1": 0.
|
| 285 |
-
"val_weighted_f1": 0.
|
| 286 |
-
},
|
| 287 |
-
{
|
| 288 |
-
"epoch": 10,
|
| 289 |
-
"train_loss": 0.4370355563504355,
|
| 290 |
-
"train_acc": 0.8125,
|
| 291 |
-
"lr_head": 3.985215271097539e-06,
|
| 292 |
-
"lr_backbone": 3.985215271097539e-07,
|
| 293 |
-
"val_loss": 0.5129753907521566,
|
| 294 |
-
"val_accuracy": 0.7333333333333333,
|
| 295 |
-
"val_macro_f1": 0.7321428571428572,
|
| 296 |
-
"val_weighted_f1": 0.7321428571428572
|
| 297 |
}
|
| 298 |
],
|
| 299 |
"stage_c": [
|
| 300 |
{
|
| 301 |
"epoch": 1,
|
| 302 |
-
"train_loss": 0.
|
| 303 |
-
"train_acc": 0.
|
| 304 |
"lr_head": 5.000000000000001e-07,
|
| 305 |
"lr_backbone": 1.5000000000000002e-07,
|
| 306 |
-
"val_loss": 0.
|
| 307 |
-
"val_accuracy": 0.
|
| 308 |
-
"val_macro_f1": 0.
|
| 309 |
-
"val_weighted_f1": 0.
|
| 310 |
},
|
| 311 |
{
|
| 312 |
"epoch": 2,
|
| 313 |
-
"train_loss": 0.
|
| 314 |
-
"train_acc": 0.
|
| 315 |
"lr_head": 5e-05,
|
| 316 |
"lr_backbone": 1.5e-05,
|
| 317 |
-
"val_loss": 0.
|
| 318 |
-
"val_accuracy": 0.
|
| 319 |
-
"val_macro_f1": 0.
|
| 320 |
-
"val_weighted_f1": 0.
|
| 321 |
},
|
| 322 |
{
|
| 323 |
"epoch": 3,
|
| 324 |
-
"train_loss": 0.
|
| 325 |
-
"train_acc": 0.
|
| 326 |
"lr_head": 4.899745109695881e-05,
|
| 327 |
"lr_backbone": 1.4699235329087644e-05,
|
| 328 |
-
"val_loss": 0.
|
| 329 |
-
"val_accuracy": 0.
|
| 330 |
-
"val_macro_f1": 0.
|
| 331 |
-
"val_weighted_f1": 0.
|
| 332 |
},
|
| 333 |
{
|
| 334 |
"epoch": 4,
|
| 335 |
-
"train_loss": 0.
|
| 336 |
-
"train_acc": 0.
|
| 337 |
"lr_head": 4.6071024937571735e-05,
|
| 338 |
"lr_backbone": 1.3821307481271522e-05,
|
| 339 |
-
"val_loss": 0.
|
| 340 |
-
"val_accuracy": 0.
|
| 341 |
-
"val_macro_f1": 0.
|
| 342 |
-
"val_weighted_f1": 0.
|
| 343 |
},
|
| 344 |
{
|
| 345 |
"epoch": 5,
|
| 346 |
-
"train_loss": 0.
|
| 347 |
-
"train_acc": 0.
|
| 348 |
"lr_head": 4.145780316514581e-05,
|
| 349 |
"lr_backbone": 1.2437340949543742e-05,
|
| 350 |
-
"val_loss": 0.
|
| 351 |
-
"val_accuracy": 0.
|
| 352 |
-
"val_macro_f1": 0.
|
| 353 |
-
"val_weighted_f1": 0.
|
| 354 |
},
|
| 355 |
{
|
| 356 |
"epoch": 6,
|
| 357 |
-
"train_loss": 0.
|
| 358 |
-
"train_acc": 0.
|
| 359 |
"lr_head": 3.5531521571796694e-05,
|
| 360 |
"lr_backbone": 1.0659456471539008e-05,
|
| 361 |
-
"val_loss": 0.
|
| 362 |
-
"val_accuracy": 0.
|
| 363 |
-
"val_macro_f1": 0.
|
| 364 |
-
"val_weighted_f1": 0.
|
| 365 |
},
|
| 366 |
{
|
| 367 |
"epoch": 7,
|
| 368 |
-
"train_loss": 0.
|
| 369 |
-
"train_acc": 0.
|
| 370 |
"lr_head": 2.877229224726381e-05,
|
| 371 |
"lr_backbone": 8.631687674179142e-06,
|
| 372 |
-
"val_loss": 0.
|
| 373 |
-
"val_accuracy": 0.
|
| 374 |
-
"val_macro_f1": 0.
|
| 375 |
-
"val_weighted_f1": 0.
|
| 376 |
-
},
|
| 377 |
-
{
|
| 378 |
-
"epoch": 8,
|
| 379 |
-
"train_loss": 0.22606890329292842,
|
| 380 |
-
"train_acc": 0.9553571428571429,
|
| 381 |
-
"lr_head": 2.1727707752736196e-05,
|
| 382 |
-
"lr_backbone": 6.518312325820858e-06,
|
| 383 |
-
"val_loss": 0.4554001530011495,
|
| 384 |
-
"val_accuracy": 0.7833333333333333,
|
| 385 |
-
"val_macro_f1": 0.783273131425396,
|
| 386 |
-
"val_weighted_f1": 0.783273131425396
|
| 387 |
-
},
|
| 388 |
-
{
|
| 389 |
-
"epoch": 9,
|
| 390 |
-
"train_loss": 0.19964813192685446,
|
| 391 |
-
"train_acc": 0.9672619047619048,
|
| 392 |
-
"lr_head": 1.4968478428203314e-05,
|
| 393 |
-
"lr_backbone": 4.490543528460994e-06,
|
| 394 |
-
"val_loss": 0.45547234614690146,
|
| 395 |
-
"val_accuracy": 0.8333333333333334,
|
| 396 |
-
"val_macro_f1": 0.8325892857142857,
|
| 397 |
-
"val_weighted_f1": 0.8325892857142857
|
| 398 |
-
},
|
| 399 |
-
{
|
| 400 |
-
"epoch": 10,
|
| 401 |
-
"train_loss": 0.21034001807371774,
|
| 402 |
-
"train_acc": 0.9583333333333334,
|
| 403 |
-
"lr_head": 9.042196834854196e-06,
|
| 404 |
-
"lr_backbone": 2.712659050456259e-06,
|
| 405 |
-
"val_loss": 0.45674578348795575,
|
| 406 |
-
"val_accuracy": 0.8333333333333334,
|
| 407 |
-
"val_macro_f1": 0.8325892857142857,
|
| 408 |
-
"val_weighted_f1": 0.8325892857142857
|
| 409 |
-
},
|
| 410 |
-
{
|
| 411 |
-
"epoch": 11,
|
| 412 |
-
"train_loss": 0.1949910166717711,
|
| 413 |
-
"train_acc": 0.9702380952380952,
|
| 414 |
-
"lr_head": 4.428975062428262e-06,
|
| 415 |
-
"lr_backbone": 1.3286925187284787e-06,
|
| 416 |
-
"val_loss": 0.4595174193382263,
|
| 417 |
-
"val_accuracy": 0.8166666666666667,
|
| 418 |
-
"val_macro_f1": 0.8153846153846154,
|
| 419 |
-
"val_weighted_f1": 0.8153846153846154
|
| 420 |
-
},
|
| 421 |
-
{
|
| 422 |
-
"epoch": 12,
|
| 423 |
-
"train_loss": 0.1842061841771716,
|
| 424 |
-
"train_acc": 0.9761904761904762,
|
| 425 |
-
"lr_head": 1.502548903041193e-06,
|
| 426 |
-
"lr_backbone": 4.5076467091235787e-07,
|
| 427 |
-
"val_loss": 0.4608180999755859,
|
| 428 |
-
"val_accuracy": 0.8166666666666667,
|
| 429 |
-
"val_macro_f1": 0.8153846153846154,
|
| 430 |
-
"val_weighted_f1": 0.8153846153846154
|
| 431 |
}
|
| 432 |
]
|
| 433 |
},
|
| 434 |
"confusion_matrix": [
|
| 435 |
[
|
| 436 |
-
|
| 437 |
-
|
| 438 |
],
|
| 439 |
[
|
| 440 |
-
|
| 441 |
-
|
| 442 |
]
|
| 443 |
],
|
| 444 |
-
"report": " precision recall f1-score support\n\n Gyuyig 0.
|
| 445 |
"idx_to_label": {
|
| 446 |
"0": "Gyuyig",
|
| 447 |
"1": "Tsugdri"
|
|
|
|
| 1 |
{
|
| 2 |
+
"experiment": "dinov3_gyuyig_tsugdri_binary",
|
| 3 |
"run_subdir": null,
|
| 4 |
"task": "gyuyig_tsugdri_binary_classification",
|
| 5 |
"balanced_parquet_dir": null,
|
|
|
|
| 8 |
"benchmark_dataset_repo": null,
|
| 9 |
"benchmark_per_parent": 60,
|
| 10 |
"preprocess": {
|
| 11 |
+
"train": "center_crop",
|
| 12 |
+
"val": "center_crop",
|
| 13 |
+
"test": "center_crop",
|
| 14 |
+
"size": 224
|
| 15 |
},
|
| 16 |
"pooling": "cls_token",
|
| 17 |
"training_config": {
|
|
|
|
| 60 |
"warmstart_checkpoint": null,
|
| 61 |
"warmstart_checkpoint_file": "final_model.pt",
|
| 62 |
"checkpoint_test_metrics": {
|
| 63 |
+
"loss": 0.42663901050885517,
|
| 64 |
+
"accuracy": 0.9208333333333333,
|
| 65 |
+
"macro_f1": 0.9208489161207983,
|
| 66 |
+
"weighted_f1": 0.9208489161207983
|
| 67 |
},
|
| 68 |
"warmstart_pooling": "cls_token"
|
| 69 |
},
|
|
|
|
| 72 |
},
|
| 73 |
"best_checkpoint": {
|
| 74 |
"path": "best_stage_c_last_blocks.pt",
|
| 75 |
+
"epoch": 1,
|
| 76 |
+
"val_macro_f1": 0.9164578111946533
|
| 77 |
},
|
| 78 |
"val_metrics": {
|
| 79 |
+
"loss": 0.3914561231931051,
|
| 80 |
+
"accuracy": 0.9166666666666666,
|
| 81 |
+
"macro_f1": 0.9164578111946533,
|
| 82 |
+
"weighted_f1": 0.9164578111946533,
|
| 83 |
+
"auc_roc": 0.9311111111111111
|
| 84 |
},
|
| 85 |
"val_confusion_matrix": [
|
| 86 |
[
|
| 87 |
+
29,
|
| 88 |
+
1
|
| 89 |
],
|
| 90 |
[
|
| 91 |
+
4,
|
| 92 |
+
26
|
| 93 |
]
|
| 94 |
],
|
| 95 |
+
"val_report": " precision recall f1-score support\n\n Gyuyig 0.88 0.97 0.92 30\n Tsugdri 0.96 0.87 0.91 30\n\n accuracy 0.92 60\n macro avg 0.92 0.92 0.92 60\nweighted avg 0.92 0.92 0.92 60\n",
|
| 96 |
"test_metrics": {
|
| 97 |
+
"loss": 0.40474860469500223,
|
| 98 |
+
"accuracy": 0.85,
|
| 99 |
+
"macro_f1": 0.847930160518164,
|
| 100 |
+
"weighted_f1": 0.847930160518164,
|
| 101 |
+
"auc_roc": 0.9297222222222223
|
| 102 |
},
|
| 103 |
"test_confusion_matrix": [
|
| 104 |
[
|
| 105 |
+
58,
|
| 106 |
+
2
|
| 107 |
],
|
| 108 |
[
|
| 109 |
+
16,
|
| 110 |
+
44
|
| 111 |
]
|
| 112 |
],
|
| 113 |
+
"test_report": " precision recall f1-score support\n\n Gyuyig 0.78 0.97 0.87 60\n Tsugdri 0.96 0.73 0.83 60\n\n accuracy 0.85 120\n macro avg 0.87 0.85 0.85 120\nweighted avg 0.87 0.85 0.85 120\n",
|
| 114 |
"history": {
|
| 115 |
"stage_a": [
|
| 116 |
{
|
| 117 |
"epoch": 1,
|
| 118 |
+
"train_loss": 0.8516533403169542,
|
| 119 |
+
"train_acc": 0.4017857142857143,
|
| 120 |
"lr_head": 5e-06,
|
| 121 |
+
"val_loss": 0.8099705815315247,
|
| 122 |
+
"val_accuracy": 0.36666666666666664,
|
| 123 |
+
"val_macro_f1": 0.36666666666666664,
|
| 124 |
+
"val_weighted_f1": 0.36666666666666664
|
| 125 |
},
|
| 126 |
{
|
| 127 |
"epoch": 2,
|
| 128 |
+
"train_loss": 0.6583722290538606,
|
| 129 |
+
"train_acc": 0.6309523809523809,
|
| 130 |
"lr_head": 0.0005,
|
| 131 |
+
"val_loss": 0.6123360713322957,
|
| 132 |
+
"val_accuracy": 0.65,
|
| 133 |
+
"val_macro_f1": 0.6378269617706237,
|
| 134 |
+
"val_weighted_f1": 0.6378269617706238
|
| 135 |
},
|
| 136 |
{
|
| 137 |
"epoch": 3,
|
| 138 |
+
"train_loss": 0.5344764121941158,
|
| 139 |
+
"train_acc": 0.7291666666666666,
|
| 140 |
"lr_head": 0.0004668412874366486,
|
| 141 |
+
"val_loss": 0.5156712194283803,
|
| 142 |
+
"val_accuracy": 0.7833333333333333,
|
| 143 |
+
"val_macro_f1": 0.7827903091060986,
|
| 144 |
+
"val_weighted_f1": 0.7827903091060985
|
| 145 |
},
|
| 146 |
{
|
| 147 |
"epoch": 4,
|
| 148 |
+
"train_loss": 0.4900844920249212,
|
| 149 |
+
"train_acc": 0.7529761904761905,
|
| 150 |
"lr_head": 0.00037624999999999996,
|
| 151 |
+
"val_loss": 0.45860581994056704,
|
| 152 |
+
"val_accuracy": 0.7666666666666667,
|
| 153 |
+
"val_macro_f1": 0.7624434389140271,
|
| 154 |
+
"val_weighted_f1": 0.7624434389140272
|
| 155 |
},
|
| 156 |
{
|
| 157 |
"epoch": 5,
|
| 158 |
+
"train_loss": 0.4637424250443776,
|
| 159 |
+
"train_acc": 0.7857142857142857,
|
| 160 |
"lr_head": 0.0002525,
|
| 161 |
+
"val_loss": 0.43475709557533265,
|
| 162 |
+
"val_accuracy": 0.7666666666666667,
|
| 163 |
+
"val_macro_f1": 0.7624434389140271,
|
| 164 |
+
"val_weighted_f1": 0.7624434389140272
|
| 165 |
},
|
| 166 |
{
|
| 167 |
"epoch": 6,
|
| 168 |
+
"train_loss": 0.4428629179795583,
|
| 169 |
+
"train_acc": 0.7827380952380952,
|
| 170 |
"lr_head": 0.00012875000000000007,
|
| 171 |
+
"val_loss": 0.4212049206097921,
|
| 172 |
+
"val_accuracy": 0.85,
|
| 173 |
+
"val_macro_f1": 0.8499583217560434,
|
| 174 |
+
"val_weighted_f1": 0.8499583217560432
|
| 175 |
},
|
| 176 |
{
|
| 177 |
"epoch": 7,
|
| 178 |
+
"train_loss": 0.4315942114307767,
|
| 179 |
+
"train_acc": 0.8095238095238095,
|
| 180 |
"lr_head": 3.815871256335142e-05,
|
| 181 |
+
"val_loss": 0.4180481950441996,
|
| 182 |
+
"val_accuracy": 0.85,
|
| 183 |
+
"val_macro_f1": 0.8499583217560434,
|
| 184 |
+
"val_weighted_f1": 0.8499583217560432
|
| 185 |
}
|
| 186 |
],
|
| 187 |
"stage_b": [
|
| 188 |
{
|
| 189 |
"epoch": 1,
|
| 190 |
+
"train_loss": 0.42322553339458646,
|
| 191 |
+
"train_acc": 0.8125,
|
| 192 |
"lr_head": 1.0000000000000002e-06,
|
| 193 |
"lr_backbone": 1.0000000000000001e-07,
|
| 194 |
+
"val_loss": 0.42059502998987836,
|
| 195 |
+
"val_accuracy": 0.8666666666666667,
|
| 196 |
+
"val_macro_f1": 0.8666666666666667,
|
| 197 |
+
"val_weighted_f1": 0.8666666666666667
|
| 198 |
},
|
| 199 |
{
|
| 200 |
"epoch": 2,
|
| 201 |
+
"train_loss": 0.43337753273191904,
|
| 202 |
+
"train_acc": 0.8095238095238095,
|
| 203 |
"lr_head": 0.0001,
|
| 204 |
"lr_backbone": 1e-05,
|
| 205 |
+
"val_loss": 0.4085062007109324,
|
| 206 |
+
"val_accuracy": 0.8666666666666667,
|
| 207 |
+
"val_macro_f1": 0.8666666666666667,
|
| 208 |
+
"val_weighted_f1": 0.8666666666666667
|
| 209 |
},
|
| 210 |
{
|
| 211 |
"epoch": 3,
|
| 212 |
+
"train_loss": 0.4347181845278967,
|
| 213 |
+
"train_acc": 0.7886904761904762,
|
| 214 |
"lr_head": 9.701478472890248e-05,
|
| 215 |
"lr_backbone": 9.701478472890248e-06,
|
| 216 |
+
"val_loss": 0.3922509431838989,
|
| 217 |
+
"val_accuracy": 0.9166666666666666,
|
| 218 |
+
"val_macro_f1": 0.9164578111946533,
|
| 219 |
+
"val_weighted_f1": 0.9164578111946533
|
| 220 |
},
|
| 221 |
{
|
| 222 |
"epoch": 4,
|
| 223 |
+
"train_loss": 0.4287990019434974,
|
| 224 |
+
"train_acc": 0.7708333333333334,
|
| 225 |
"lr_head": 8.84191999343894e-05,
|
| 226 |
"lr_backbone": 8.841919993438941e-06,
|
| 227 |
+
"val_loss": 0.3846385677655538,
|
| 228 |
+
"val_accuracy": 0.85,
|
| 229 |
+
"val_macro_f1": 0.849624060150376,
|
| 230 |
+
"val_weighted_f1": 0.849624060150376
|
| 231 |
},
|
| 232 |
{
|
| 233 |
"epoch": 5,
|
| 234 |
+
"train_loss": 0.40112980774470736,
|
| 235 |
+
"train_acc": 0.8214285714285714,
|
| 236 |
"lr_head": 7.525e-05,
|
| 237 |
"lr_backbone": 7.525e-06,
|
| 238 |
+
"val_loss": 0.3696295181910197,
|
| 239 |
+
"val_accuracy": 0.9,
|
| 240 |
+
"val_macro_f1": 0.899888765294772,
|
| 241 |
+
"val_weighted_f1": 0.899888765294772
|
| 242 |
},
|
| 243 |
{
|
| 244 |
"epoch": 6,
|
| 245 |
+
"train_loss": 0.3798623964900062,
|
| 246 |
+
"train_acc": 0.8422619047619048,
|
| 247 |
"lr_head": 5.909558479451306e-05,
|
| 248 |
"lr_backbone": 5.909558479451306e-06,
|
| 249 |
+
"val_loss": 0.365932967265447,
|
| 250 |
+
"val_accuracy": 0.9,
|
| 251 |
+
"val_macro_f1": 0.899888765294772,
|
| 252 |
+
"val_weighted_f1": 0.899888765294772
|
| 253 |
},
|
| 254 |
{
|
| 255 |
"epoch": 7,
|
| 256 |
+
"train_loss": 0.3871297893070039,
|
| 257 |
+
"train_acc": 0.8214285714285714,
|
| 258 |
"lr_head": 4.190441520548695e-05,
|
| 259 |
"lr_backbone": 4.190441520548696e-06,
|
| 260 |
+
"val_loss": 0.3574748694896698,
|
| 261 |
+
"val_accuracy": 0.8833333333333333,
|
| 262 |
+
"val_macro_f1": 0.883300916921367,
|
| 263 |
+
"val_weighted_f1": 0.883300916921367
|
| 264 |
},
|
| 265 |
{
|
| 266 |
"epoch": 8,
|
| 267 |
+
"train_loss": 0.39022268142019,
|
| 268 |
+
"train_acc": 0.8005952380952381,
|
| 269 |
"lr_head": 2.5750000000000013e-05,
|
| 270 |
"lr_backbone": 2.575000000000001e-06,
|
| 271 |
+
"val_loss": 0.3670339067776998,
|
| 272 |
+
"val_accuracy": 0.85,
|
| 273 |
+
"val_macro_f1": 0.849624060150376,
|
| 274 |
+
"val_weighted_f1": 0.849624060150376
|
| 275 |
},
|
| 276 |
{
|
| 277 |
"epoch": 9,
|
| 278 |
+
"train_loss": 0.3902270041760944,
|
| 279 |
+
"train_acc": 0.7976190476190477,
|
| 280 |
"lr_head": 1.2580800065610596e-05,
|
| 281 |
"lr_backbone": 1.2580800065610596e-06,
|
| 282 |
+
"val_loss": 0.3644895474116007,
|
| 283 |
+
"val_accuracy": 0.8666666666666667,
|
| 284 |
+
"val_macro_f1": 0.8665183537263625,
|
| 285 |
+
"val_weighted_f1": 0.8665183537263625
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 286 |
}
|
| 287 |
],
|
| 288 |
"stage_c": [
|
| 289 |
{
|
| 290 |
"epoch": 1,
|
| 291 |
+
"train_loss": 0.4103491987500872,
|
| 292 |
+
"train_acc": 0.8125,
|
| 293 |
"lr_head": 5.000000000000001e-07,
|
| 294 |
"lr_backbone": 1.5000000000000002e-07,
|
| 295 |
+
"val_loss": 0.3914561231931051,
|
| 296 |
+
"val_accuracy": 0.9166666666666666,
|
| 297 |
+
"val_macro_f1": 0.9164578111946533,
|
| 298 |
+
"val_weighted_f1": 0.9164578111946533
|
| 299 |
},
|
| 300 |
{
|
| 301 |
"epoch": 2,
|
| 302 |
+
"train_loss": 0.4436623099304381,
|
| 303 |
+
"train_acc": 0.7678571428571429,
|
| 304 |
"lr_head": 5e-05,
|
| 305 |
"lr_backbone": 1.5e-05,
|
| 306 |
+
"val_loss": 0.3777097463607788,
|
| 307 |
+
"val_accuracy": 0.9166666666666666,
|
| 308 |
+
"val_macro_f1": 0.9164578111946533,
|
| 309 |
+
"val_weighted_f1": 0.9164578111946533
|
| 310 |
},
|
| 311 |
{
|
| 312 |
"epoch": 3,
|
| 313 |
+
"train_loss": 0.3808557249250866,
|
| 314 |
+
"train_acc": 0.8363095238095238,
|
| 315 |
"lr_head": 4.899745109695881e-05,
|
| 316 |
"lr_backbone": 1.4699235329087644e-05,
|
| 317 |
+
"val_loss": 0.3577085534731547,
|
| 318 |
+
"val_accuracy": 0.8666666666666667,
|
| 319 |
+
"val_macro_f1": 0.8665183537263625,
|
| 320 |
+
"val_weighted_f1": 0.8665183537263625
|
| 321 |
},
|
| 322 |
{
|
| 323 |
"epoch": 4,
|
| 324 |
+
"train_loss": 0.3782187302907308,
|
| 325 |
+
"train_acc": 0.8392857142857143,
|
| 326 |
"lr_head": 4.6071024937571735e-05,
|
| 327 |
"lr_backbone": 1.3821307481271522e-05,
|
| 328 |
+
"val_loss": 0.3452535013357798,
|
| 329 |
+
"val_accuracy": 0.8833333333333333,
|
| 330 |
+
"val_macro_f1": 0.883300916921367,
|
| 331 |
+
"val_weighted_f1": 0.883300916921367
|
| 332 |
},
|
| 333 |
{
|
| 334 |
"epoch": 5,
|
| 335 |
+
"train_loss": 0.38378763056936716,
|
| 336 |
+
"train_acc": 0.8392857142857143,
|
| 337 |
"lr_head": 4.145780316514581e-05,
|
| 338 |
"lr_backbone": 1.2437340949543742e-05,
|
| 339 |
+
"val_loss": 0.33214412728945414,
|
| 340 |
+
"val_accuracy": 0.8833333333333333,
|
| 341 |
+
"val_macro_f1": 0.883300916921367,
|
| 342 |
+
"val_weighted_f1": 0.883300916921367
|
| 343 |
},
|
| 344 |
{
|
| 345 |
"epoch": 6,
|
| 346 |
+
"train_loss": 0.3539558429093588,
|
| 347 |
+
"train_acc": 0.8571428571428571,
|
| 348 |
"lr_head": 3.5531521571796694e-05,
|
| 349 |
"lr_backbone": 1.0659456471539008e-05,
|
| 350 |
+
"val_loss": 0.3277231236298879,
|
| 351 |
+
"val_accuracy": 0.8833333333333333,
|
| 352 |
+
"val_macro_f1": 0.883300916921367,
|
| 353 |
+
"val_weighted_f1": 0.883300916921367
|
| 354 |
},
|
| 355 |
{
|
| 356 |
"epoch": 7,
|
| 357 |
+
"train_loss": 0.35197025750364574,
|
| 358 |
+
"train_acc": 0.8511904761904762,
|
| 359 |
"lr_head": 2.877229224726381e-05,
|
| 360 |
"lr_backbone": 8.631687674179142e-06,
|
| 361 |
+
"val_loss": 0.3296816150347392,
|
| 362 |
+
"val_accuracy": 0.8666666666666667,
|
| 363 |
+
"val_macro_f1": 0.8665183537263625,
|
| 364 |
+
"val_weighted_f1": 0.8665183537263625
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 365 |
}
|
| 366 |
]
|
| 367 |
},
|
| 368 |
"confusion_matrix": [
|
| 369 |
[
|
| 370 |
+
58,
|
| 371 |
+
2
|
| 372 |
],
|
| 373 |
[
|
| 374 |
+
16,
|
| 375 |
+
44
|
| 376 |
]
|
| 377 |
],
|
| 378 |
+
"report": " precision recall f1-score support\n\n Gyuyig 0.78 0.97 0.87 60\n Tsugdri 0.96 0.73 0.83 60\n\n accuracy 0.85 120\n macro avg 0.87 0.85 0.85 120\nweighted avg 0.87 0.85 0.85 120\n",
|
| 379 |
"idx_to_label": {
|
| 380 |
"0": "Gyuyig",
|
| 381 |
"1": "Tsugdri"
|
training_history.png
CHANGED
|
|